Search NASA⌕ Search

SEARCH · Search NASA

Results for “ITERATIVE NETWORK”

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 109 records · Page 6
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Stochastic machine learning via sigma profiles to build a digital chemical space

This work establishes a different paradigm on digital molecular spaces and their efficient navigation by exploiting sigma profiles. To do so, the remarkable capability of Gaussian processes (GPs), a type of stochastic machine learning model, to correlate and predict physicochemical properties from sigma profiles is demonstrated, outperforming state-of-the-art neural networks previously published. The amount of chemical information encoded in sigma profiles eases the learning burden of machine learning models, permitting the training of GPs on small datasets which, due to their negligible computational cost and ease of implementation, are ideal models to be combined with optimization tools such as gradient search or Bayesian optimization (BO). Gradient search is used to efficiently navigate the sigma profile digital space, quickly converging to local extrema of target physicochemical properties. While this requires the availability of pretrained GP models on existing datasets, such limitations are eliminated with the implementation of BO, which can find global extrema with a limited number of iterations. A remarkable example of this is that of BO toward boiling temperature optimization. Holding no knowledge of chemistry except for the sigma profile and boiling temperature of carbon monoxide (the worst possible initial guess), BO finds the global maximum of the available boiling temperature dataset (over 1,000 molecules encompassing more than 40 families of organic and inorganic compounds) in just 15 iterations (i.e., 15 property measurements), cementing sigma profiles as a powerful digital chemical space for molecular optimization and discovery, particularly when little to no experimental data is initially available.

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

Applications of Lifted Nonlinear Cuts to Convex Relaxations of the AC Power Flow Equations

Here, we demonstrate that valid inequalities, or lifted nonlinear cuts (LNC), can be projected to tighten the Second Order Cone (SOC), Convex DistFlow (CDF), and Network Flow (NF) relaxations of the AC Optimal Power Flow (AC-OPF) problem. We conduct experiments on 38 cases from the PGLib-OPF library, showing that the LNC strengthen the SOC and CDF relaxations in 100% of the test cases, with average and maximum differences in the optimality gaps of 6.2% and 17.5% respectively. The NF relaxation is strengthened in 46.2% of test cases, with average and maximum differences in the optimality gaps of 1.3% and 17.3% respectively. We also study the trade-off between relaxation quality and solve time, demonstrating that the strengthened CDF relaxation outperforms the strengthened SOC formulation in terms of runtime and number of iterations needed, while the strengthened NF formulation is the most scalable with the lowest relaxation quality improvement due to these LNC.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2025

Active deep kernel learning of molecular properties from structural embeddings

As vast databases of chemical identities become increasingly available, the challenge shifts to how we effectively explore and leverage these resources to study molecular properties. This paper presents an active learning approach for molecular discovery using deep kernel learning (DKL), demonstrated on the QM9 dataset. DKL links structural embeddings directly to properties, creating organized latent spaces that prioritize relevant property information. By iteratively recalculating embedding vectors in alignment with target properties, DKL uncovers concentrated maxima representing key molecular properties and reveals unexplored regions with potential for innovation. This approach underscores DKL’s potential in advancing molecular research and discovery.

Artificial neural networks↗
DOE OSTI2025

Fast meta-solvers for 3D complex-shape scatterers using neural operators trained on a non-scattering problem

Three-dimensional target identification using scattering techniques requires high accuracy solutions and very fast computations for real-time predictions in some critical applications. We first train a deep neural operator (DeepONet) to solve wave propagation problems described by the Helmholtz equation in a domain without scatterers but at different wavenumbers and with a complex absorbing boundary condition. We then design two classes of fast meta-solvers by combining DeepONet with either relaxation methods, such as Jacobi and Gauss-Seidel, or with Krylov methods, such as GMRES and BiCGStab, using the trunk basis of DeepONet as a coarse-scale preconditioner. We leverage the spectral bias of neural networks to account for the lower part of the spectrum in the error distribution while the upper part is handled inexpensively using relaxation methods or fine-scale preconditioners. The meta-solvers are then applied to solve scattering problems with different shape of scatterers, at no extra training cost. We first demonstrate that the resulting meta-solvers are shape-agnostic, fast, and robust, whereas the standard standalone solvers may even fail to converge without the DeepONet. We then apply both classes of meta-solvers to scattering from a submarine, a complex three-dimensional problem. We achieve very fast solutions, especially with the DeepONet-Krylov methods, which require orders of magnitude fewer iterations than any of the standalone solvers.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

A data-driven approach to real-time vertical position estimation for NSTX-U vertical stability control

In this paper, a database of 77 996 plasma equilibrium reconstructions from 727 discharges during the initial operation of the NSTX-U spherical tokamak is analyzed to develop a statistically robust model of the plasma vertical position for real-time control. A variety of regression models are developed and tested, ranging in complexity from linear models to deep neural networks, and including input signals ranging from the four pairs of flux loops used historically on NSTX-U up to the full set of 389 real-time signals available to the plasma control system. A linear model based on 140 real-time magnetics signals is found to offer excellent accuracy, with a coefficient of determination R 2 = 0.906. The robustness of this model to limited training data, new operating scenarios, and signal errors is tested, and a procedure is demonstrated to tune the model parameters to optimize its robustness. A time-dependent plasma equilibrium solver, TokaMaker, is used to simulate vertical stability control in NSTX-U, demonstrating that it should be possible to iteratively tune the parameters of a linear vertical position model to stabilize both positive and negative triangularity plasmas in future experiments.

magnetic diagnostics↗
DOE OSTI2026

SODAs: sparse optimization for the discovery of differential and algebraic equations

Differential-algebraic equations (DAEs) integrate ordinary differential equations (ODEs) with algebraic constraints, providing a fundamental framework for developing models of dynamical systems characterized by time-scale separation, conservation laws and physical constraints. While sparse optimization has revolutionized model development by allowing data-driven discovery of parsimonious models from a library of possible equations, existing approaches for dynamical systems assume DAEs can be reduced to ODEs by eliminating variables before model discovery. This assumption limits the applicability of such methods for DAE systems with unknown constraints and time scales. We introduce sparse optimization for differential-algebraic systems (SODAs), a data-driven method for the identification of DAEs in their explicit form. By discovering the algebraic and dynamic components sequentially without prior identification of the algebraic variables, this approach leads to a sequence of convex optimization problems. It has the advantage of discovering interpretable models that preserve the structure of the underlying physical system. To this end, SODAs improves since SODAs is singular numerical stability when handling high correlations between library terms, caused by near-perfect algebraic relationships, by iteratively refining the conditioning of the candidate library. We demonstrate the performance of our method on biological, mechanical and electrical systems, showcasing its robustness to noise in both simulated time series and real-time experimental data.

DAE↗
DOE OSTI2024

A Science Gateway for the Repeatable Analysis of Machine Learning Predicted Gravity Anomalies

In recent years, deep learning has become an increasingly popular alternative for modeling in geoscience applications due to its scalability and efficiency. However, the interpretability, compute, data volume, and hyperparameter tuning requirements of deep learning models make development and monitoring difficult. Furthermore, model explainability and communicating results obtained by these models to users or domain experts is a challenge, as domain experts in geoscience also need to have a deep understanding of how those models function in order to support their scientific works. Here, we describe a science gateway and machine learning pipeline for predicting gravity anomalies from geophysical data. The gateway, built on open-source technologies, provides a holistic view of the pipeline through interactive visualizations aimed at enabling efficient exploratory data analysis. The repeatability, reproducibility, and monitoring capabilities of this overall system allow us to iterate and analyze at scale. Using this pipeline and gateway, we can repeatedly produce accurate high-resolution gravity anomaly datasets. By describing the underlying technologies, implementation, and results, here we provide a foundation for the broader adoption of science gateways into cross-cutting geoscience and machine learning research projects as a means to improve the scientific discovery and collaboration in the geophysics and computational sciences community.

58 GEOSCIENCES↗
DOE OSTI2025

Event Classifications on DNE2 Main Experiment Data using a Convolutional Neural Network Ensemble

The Dynamic Networks (DN) Experiment for FY24 (DNE2) is an experiment within DN with the goal of quantitatively evaluating the effectiveness of solutions developed so far by various researchers under the Low Yield Nuclear Monitoring (LYNM) program using a shared set of metrics and datasets. A key component of this experiment is the mimicking of a signature processing pipeline, and comparing currently accepted and standard-use processing methods to more state-of-the-art processes developed under DN. In this work, we focus specifically on the Event Characterization (EC) Focus Area (FA) of the pipeline, where a seismic event’s magnitude, yield and class are identified. We use Deep Learning (DL) to classify the type of events being processed as either earthquakes (EQs) or explosions (EXs) for three iterations of experiment datasets. The model is noticeably more confident and accurate in classifying explosions than earthquakes, reflecting a known shortcoming of the model, that being of a bias towards predicting explosions over earthquakes in the west coast due to training data biases.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

A Kaczmarz-inspired approach to accelerate the optimization of neural network wavefunctions

Neural network wavefunctions optimized using the variational Monte Carlo method have been shown to produce highly accurate results for the electronic structure of atoms and small molecules, but the high cost of optimizing such wavefunctions prevents their application to larger systems. We propose the Subsampled Projected-Increment Natural Gradient Descent (SPRING) optimizer to reduce this bottleneck. SPRING combines ideas from the recently introduced minimum-step stochastic reconfiguration optimizer (MinSR) and the classical randomized Kaczmarz method for solving linear least-squares problems. We demonstrate that SPRING outperforms both MinSR and the popular Kronecker-Factored Approximate Curvature method (KFAC) across a number of small atoms and molecules, given that the learning rates of all methods are optimally tuned. For example, on the oxygen atom, SPRING attains chemical accuracy after forty thousand training iterations, whereas both MinSR and KFAC fail to do so even after one hundred thousand iterations.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

KIPM Detector Characterization in QUIET

Kinetic Inductance Phonon‑Mediated (KIPM) detectors are superconducting microwave resonators that sense energy depositions in the substrate through a transient shift in their resonant frequency. In this work, I installed new devices in the QUIET cryogenic facility and used a network analyzer to drive them across a range of power levels. At each setting, I measured the transmitted signal and fit it to a simple notch‑filter model, allowing us to extract key performance metrics. Despite the increased noise at low power, the model consistently captured the resonance behavior. I first measured the resonance parameters as a function of RF drive power, then repeated the measurement across a range of temperatures to test whether the device behaved like an MKID. While the notch-filter fits captured the resonance reliably, the observed temperature dependence did not match the expected MKID signature — indicating that these first in-house fabricated chips are not yet functioning as true MKIDs. Overall, this reframes our work as a study of resonance parameter extraction and identification, pointing towards future iterations of device design and fabrication to get the true MKID response.

Savitala, Akash [U. Washington, Seattle (main); Fe↗
DOE OSTI2026

CHESS 2025: Leaf Area Index (LAI) for meadow, shrub, tree, and understory vegetation

This dataset contains Leaf Area Index (LAI) measurements made as part of the Colorado Headwaters Ecological Spectroscopy Study (CHESS) during June and July of 2025. Data were collected in the Upper Gunnison Basin, Colorado, across three study domains: the Upper East River (CRBU), Almont Triangle (ALMO), and the Upper Taylor Basin (UPTA). Field observations of LAI were collected within 72 hours of airborne data collection by the National Ecological Observatory Network’s Aerial Observation Platform (NEON AOP). The NEON AOP collected waveform LiDAR (Light Detection and Ranging) and imaging spectrometer data in 426 spectral bands from the visible to shortwave infrared. LAI measurements were collected using the LICOR LAI-2200C Plant Canopy Analyzer following protocols outlined in the instrument manual (LI-COR 2019). Sampling targeted four distinct vegetation types: meadows, shrubs, trees, and aspen forest understory. We have archived data separately by site type because different field methods were used for each. At meadow sites, measurements were made at the four corners of 1m x 1m plots, with the instrument moving inward toward the center of the plot. At shrub sites, we measured the canopies of individual shrubs. At tree sites, we made measurements within a 10m x 10m subplot centered around a focal tree, with 30 observations taken on a regular grid. At aspen understory sites, we measured overstory trees following the tree protocol and understory herbaceous vegetation following the meadow protocol. All measurements included above-canopy (A) and below-canopy (B) readings, with specific protocols for scattering correction measurements in direct-sun conditions. Data were processed using the R package `rlai` (Worsham 2025). This package includes functions to calculate LAI, gap fraction, apparent clumping factor (Ω), scattering correction, and other canopy metrics. Package contents: Full file descriptions appear in ‘flmd.csv’. Files named according to the convention ‘lai_*_summary_data_cleaned.csv’ contain summary values of LAI, apparent clumping factor (Ωapp), and scattering correction factors for each site. These are the analysis-ready products that most data users will work with. Files named ‘lai_*_metadata_cleaned.csv’ contain additional site-level observations made during field collection. We have also archived intermediate and supplementary data for users who wish to check our processing approach or apply alternative methods. ‘raw_lai_2200C.zip’ contains the raw files as read from the LI-COR instrument, with no processing applied, in TXT format. The zip archive contains subdirectories by site type, which are further subdivided by sampling area. Filenames correspond to the sampling site number. ‘intermediate_results.zip’ contains detailed output from the processing routines, in JSON format. The zip archive contains subdirectories by site type; filenames correspond to the sampling site number. ‘scattering_correction_logs.zip’ contains logfiles from the implementation of Kobayashi et al.'s (2013) scattering correction algorithm. The logfiles report values of several parameters at each iteration of the algorithm, as the model converges toward a stable solution. They are intended for users who want to verify scattering correction performance. The zip archive contains subdirectories by site type; filenames correspond to the sampling site number. ‘spot_checks.csv’ reports LAI and other values for a small number of files processed with LI-COR FV2200 software (LI-COR 2013) using the same control parameters as in our R-based approach. Additional metadata are provided in a data dictionary describing column names and definitions (dd.csv), and in a file-level metadata file (flmd.csv). All zip files can be expanded with common archive utilities. TXT, CSV, and JSON files can be ingested into R or Python computing environments or read in common text editor utilities. Geospatial information: Geospatial data for mapping measurement site locations are in the files CHESS_polygons_lai_UTM.geojson, CHESS_polygons_shrub_UTM.geojson, and CHESS_polygons_meadow_UTM.geojson in the companion geospatial package for the 2025 CHESS campaign, ‘CHESS 2025: Location data for field observations and sampling’ (Henderson et al., 2026). CHESS Project Description: The Colorado Headwaters Ecological Spectroscopy Study (CHESS) comprised a multi-week airborne remote sensing and field observation campaign in the Upper Gunnison Basin, Colorado, conducted in June and July of 2025. Airborne remote sensing was conducted by the National Ecological Observatory Network Airborne Observation Platform (NEON AOP), concurrent with a field campaign run by the Rocky Mountain Biological Laboratory (RMBL), the Lawrence Berkeley National Laboratory (LBNL) and SLAC National Accelerator Laboratory Watershed Function Science Focus Area (SFA), and NASA-JPL (Jet Propulsion Laboratory) Earth Surface Mineral Dust Source Investigation (EMIT) program. Between June 10 and July 18, 2025, the NEON AOP flight team collected high-resolution aerial imaging spectroscopy and Light Detection and Ranging (LiDAR) data over three domains: the Upper East River (CRBU), Almont Triangle (ALMO), and the Upper Taylor Basin (UPTA). In coordination with the flights, a field campaign acquired ground-truth observations, including observations of vegetation composition, foliar traits, forest demography, and subsurface properties in 18 core sampling areas within the domains. Additional surface water observations were taken at over 380 point locations. All CHESS campaign datasets can be found within the CHESS ESS-DIVE data portal: https://data.ess-dive.lbl.gov/portals/chess. Funding Acknowledgement: Field and remote-sensing data acquisition was performed under a grant from the National Aeronautics and Space Administration (80NSSC24K1005). This work was also supported by the Watershed Function Science Focus Area at Lawrence Berkeley National Laboratory funded by the US Department of Energy, Office of Science, Biological and Environmental Research under Contract No. DE-AC02-05CH11231. * Todorov and Worsham are co–first authors.

2018 NEON and 2025 CHESS Campaigns↗
DOE OSTI2024

Artificial-intelligence-driven shot reduction in quantum measurement

Variational Quantum Eigensolver (VQE) provides a powerful solution for approximating molecular ground state energies by combining quantum circuits and classical computers. However, estimating probabilistic outcomes on quantum hardware requires repeated measurements (shots), incurring significant costs as accuracy increases. Optimizing shot allocation is thus critical for improving the efficiency of VQE. Current strategies rely heavily on hand-crafted heuristics requiring extensive expert knowledge. This paper proposes a reinforcement learning (RL)-based approach that automatically learns shot assignment policies to minimize total measurement shots while achieving convergence to the minimum of the energy expectation in VQE. The RL agent assigns measurement shots across VQE optimization iterations based on the progress of the optimization. This approach reduces VQE's dependence on static heuristics and human expertise. When the RL-enabled VQE is applied to a small molecule, a shot reduction policy is learned. The policy demonstrates transferability across systems and compatibility with other wavefunction Ansätze. In addition to these specific findings, this work highlights the potential of RL for automatically discovering efficient and scalable quantum optimization strategies.

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

DS-TIDE: Harnessing Dynamical Systems for Efficient Time-Independent Differential Equation Solving

Time-Independent Differential Equations (TIDEs) are central to modeling equilibrium behavior across a wide range of scientific and engineering domains, from electrostatics to porous media flow. Conventional numerical solvers offer reliable solutions but incur significant computational costs due to fine-grained discretization and iterative procedures. Machine learning-based approaches address this by replacing iterative solving processes with one-time inference; however, their sophisticated models require extensive training resources that often exceed those of traditional solvers. Consequently, designing a TIDE solver that achieves high accuracy, broad applicability, and exceptional computational efficiency remains a fundamental challenge. In this paper, we propose DS-TIDE, a novel hardware solver that is inspired by, and subsequently leverages, the intrinsic connection between Dynamical Systems (DS) and Differential Equations (DEs) to efficiently and accurately solve TIDEs. DS-TIDE employs a CMOS-compatible DS-based processor, whose physical states evolve under carefully designed DE-driven dynamics and naturally converge to equilibrium -- the solution of the target TIDE -- within ~1µs on a ~1-watt DS-TIDE processor. To enhance expressivity, DS-TIDE incorporates Heterogeneous Dynamics with Temporal Layering (HDTL), which solves TIDEs through a three-stage DS evolution -- conditioning, solving, and decoding -- each governed by specialized dynamics. The entire evolution process is analogous to an infinitely deep neural network temporally unrolled, offering the system the capability of representing complex equations. Furthermore, DS-TIDE is equipped with an on-device DS-DE Auto-Alignment mechanism that dynamically adapts intrinsic hardware dynamics within milliseconds, effectively aligning the system’s dynamics to diverse target DEs. Experimental results across TIDEs from a wide range of scientific and engineering domains demonstrate that DS-TIDE achieves ~10^3× speedup, ~10^5× energy savings, and competitive or superior accuracy compared to state-of-the-art numerical and ML-based solvers.

Liu, Chuan↗
DOE OSTI2026

Feature Engineering and Ensemble Methods for Imbalanced ICS Intrusion Detection: Pipeline Audit and Constrained Evaluation

Industries are becoming increasingly connected and are more vulnerable to cyberattacks due to the widened attack surface. Industrial Control Systems (ICS) are among the most critical sectors that malicious actors can target, as such attacks can cause significant operational disruption and physical damage. It is imperative to detect such attacks as early as possible. This paper evaluates constraint-conditioned optimistic performance estimates for traditional ML models in ICS intrusion detection (i.e., estimates obtained under contiguous, non-shuffled temporal evaluation without test-set alteration, but with pre-split feature engineering that may introduce temporal leakage, due to dataset constraints). Our findings are threefold. First, we quantify how iterative feature engineering affects tree-based ensemble performance and examine how pipeline decisions (split strategy, sampling scope, and cleaning policy) can inflate or reduce reported IDS results under constraint-bound evaluation. Second, we compare intrinsic class-imbalance handling across ensemble models. Third, under our current pipeline constraints (including pre-split feature engineering), CatBoost achieves the best performance on Water Storage Tank (accuracy: 0.9831, class-1 F1: 0.9682), while Light- GBM achieves the best performance on Gas Pipeline (accuracy: 0.9618, class-1 F1: 0.9086).

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Development Fiber Optic Distributed System for Direct Detection of Subsurface Gases Leakages

Carbon, natural gas, and hydrogen gas storage is an emerging solution to safeguard us against pollution, support goals of negative carbon emission, and protect sources of renewable energy. Properly constructed storage wells provide a virtually impervious barrier to any unintended subsurface transmission. The ability to ensure the long-term integrity of such wells is vital to the success of any storage operation and be successful in the public eyes. Therefore, robust monitoring of any gas migration into the subsurface is highly sought. A fiber-optic distributed chemical sensor (DCS) enables monitoring of long-term well integrity along its depth, ensuring the success of any storage operation and bolsters public acceptance of the safety of the reservoir via leak early detection. The same technique can be applied to gas monitoring in pipeline networks and nuclear stockpile monitoring applications. Fiber based Raman spectroscopy enables DCS, as optical fibers can be deployed in virtually any environment and relay spectroscopic information over long distances back to the user. Hollow core fibers (HCF) make excellent DCSs as the air core of the fiber allows gas from the environment to diffuse into the core, which interacts with the laser signal that is carried in the air core. This work builds upon the previous LDRD project, Fiber Optic System for Direct Detection of Carbon Dioxide Leakage in Carbon Storage Wells (21-FS-003), in which the feasibility of Raman spectroscopy detection of Carbon Dioxide (CO2) in HCF detection was demonstrated. We mitigated the risk of this DCS technology by establishing and completing five objectives. The first objective was to model and optically characterize HCF uptake of CO2, establishing the relationship between HCF length, gas diffusion time, detectable gas concentration, and measured Raman intensity. In objective two, we developed a fiber core drilling recipe to enable additional diffusion ports in the fiber core and established a method for maintaining fiber strength and integrity post drilling. Objective three characterized the drilled fibers against the undrilled fibers, establishing the differences in the gas mechanics and optical properties and provided parameters to iterate the drilling process. In objective four, a fusion splicing technique was developed to join the HCF to conventional single-mode fibers, localizing the gas detection point at the drilled HCF hole, emulating a DCS. Lastly, objective five was the testing of the sensor in Edgar Mines at Colorado School of Mines on a CO2 pipeline with a simulated leak, to showcase the ability to detect CO2 leaks. This capstone result showed CO2 leak detection in < 10 minutes, raising the technology readiness level of HCF segments as deployable DCS.

organic↗
DOE OSTI2026

How Do You Hear a Quantum Computer Whisper?

Quantum transduction is the process of upconverting microwave quantum signals into optical signals to develop quantum networks through optical fibers outside the dilution refrigerator, enabling connections between quantum technologies on the quantum internet. In this project, upconversion is achieved by directing an optical laser and the microwave quantum signal into an electro-optic bulk crystal. The crystal is housed within a Superconducting Radio Frequency (SRF) cavity designed to maximize the overlap between the microwave field and the crystal volume. In the presence of microwaves, the refractive index of the crystal changes through the Pockels effect. This change in refractive index modifies the propagation of the optical field within the crystal, allowing the quantum information carried by the microwave field to be transferred to the optical field. This study focuses on coupling laser light from suspended waveguide chips into the whispering-gallery modes (WGMs) of the crystal to enable transduction. The coupling efficiency between the optical field in the suspended waveguide and the WGM depends on the position of the laser spot on the crystal. To address this challenge, a feedback-based algorithm is being developed to fine-tune the waveguide position so that the optical signal is coupled efficiently into the crystal. After passing through the crystal, the optical signal is detected by a photodiode connected to an oscilloscope. The algorithm evaluates the coupling quality and iteratively adjusts the waveguide position to maximize coupling efficiency. The expected outcome is an automated waveguide alignment method that improves optical coupling and enables more efficient microwave-to-optical quantum transduction.

Karanastasis, Mihael [Fermilab]↗
DOE OSTI2025

Leveraging operator learning to accelerate convergence of the preconditioned conjugate gradient method

We propose a new deflation strategy to accelerate the convergence of the preconditioned conjugate gradient (PCG) method for solving parametric large-scale linear systems of equations. Unlike traditional deflation techniques that rely on eigenvector approximations or recycled Krylov subspaces, we generate the deflation subspaces using operator learning, specifically the Deep Operator Network (DeepONet). To this aim, we introduce two complementary approaches for assembling the deflation operators. The first approach approximates near-null space vectors of the discrete PDE operator using the basis functions learned by the DeepONet. The second approach directly leverages solutions predicted by the DeepONet. To further enhance convergence, we also propose several strategies for prescribing the sparsity pattern of the deflation operator. Here, a comprehensive set of numerical experiments encompassing steady-state, time-dependent, scalar, and vector-valued problems posed on both structured and unstructured geometries is presented and demonstrates the effectiveness of the proposed DeepONet-based deflated PCG method, as well as its generalization across a wide range of model parameters and problem resolutions.

Deflation↗
DOE OSTI2025

Technical Report on Subsurface Monitoring of the Brady Hot Spring Geothermal Site, Nevada, based upon Full Waveform Inversion

Abilities to accurately characterize the subsurface in a geothermal setting is key to assess and support production. An important element of geothermal reservoir monitoring is also the ability to investigate fluid transport within fracture network. This report focuses on improving subsurface imaging and monitoring in geothermal settings using full waveform inversion based on the adjoint method and time-lapse imaging. To assess our method, we rely on a dense seismic dataset collected in 2016 at the Brady Hot Springs geothermal site in Nevada for the DOE-funded project Poroelastic Tomography by Adjoint Inverse Modeling of Data from Seismology, Geodesy, and Hydrology. This dataset captures subsurface changes across four stages of geothermal power plant operations, which involve varying rates of fluid injection and extraction. Two velocity models were previously derived from this dataset using different methods: one based on travel times and another on sweep interferometry. Our first step is to refine these models using adjoint tomography, which has been applied successfully at global and regional-scales but is less common at the reservoir-scale. Two approaches are then explored for time-lapse analysis: directly comparing refined tomographic models from different stages or backpropagating waveform differences relative to a baseline tomographic model. The main take away is that both approaches highlight similar reservoir behaviors, but the latter approach is more computationally effective in capturing small-scale changes in subsurface properties. For this work, we leverage the use of Salvus (www.mondaic.com), an end-to-end seismic imaging solution, relying on the spectral element method to compute forward and adjoint simulations, and developed by Mondaic Ltd. It includes integrated workflow management that handles waveform and metadata, launches simulations, computes waveform misfits and adjoint sources, and iterates for model updates by nonlinear optimization.

15 GEOTHERMAL ENERGY↗