Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Iteration method”

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 1,153 records · Page 64
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Connectivity‐based parcellation of normal and anatomically distorted human cerebral cortex

Abstract For over a century, neuroscientists have been working toward parcellating the human cortex into distinct neurobiological regions. Modern technologies offer many parcellation methods for healthy cortices acquired through magnetic resonance imaging. However, these methods are suboptimal for personalized neurosurgical application given that pathology and resection distort the cerebrum. We sought to overcome this problem by developing a novel connectivity‐based parcellation approach that can be applied at the single‐subject level. Utilizing normative diffusion data, we first developed a machine‐learning (ML) classifier to learn the typical structural connectivity patterns of healthy subjects. Specifically, the Glasser HCP atlas was utilized as a prior to calculate the streamline connectivity between each voxel and each parcel of the atlas. Using the resultant feature vector, we determined the parcel identity of each voxel in neurosurgical patients ( n = 40) and thereby iteratively adjusted the prior. This approach enabled us to create patient‐specific maps independent of brain shape and pathological distortion. The supervised ML classifier re‐parcellated an average of 2.65% of cortical voxels across a healthy dataset ( n = 178) and an average of 5.5% in neurosurgical patients. Our patient dataset consisted of subjects with supratentorial infiltrating gliomas operated on by the senior author who then assessed the validity and practical utility of the re‐parcellated diffusion data. We demonstrate a rapid and effective ML parcellation approach to parcellation of the human cortex during anatomical distortion. Our approach overcomes limitations of indiscriminately applying atlas‐based registration from healthy subjects by employing a voxel‐wise connectivity approach based on individual data.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2021

Mixed precision s –step Lanczos and conjugate gradient algorithms

Compared to the classical Lanczos algorithm, the s-step Lanczos variant has the potential to improve performance by asymptotically decreasing the synchronization cost per iteration. However, this comes at a price; despite being mathematically equivalent, the s-step variant may behave quite differently in finite precision, potentially exhibiting greater loss of accuracy and slower convergence relative to the classical algorithm. It has previously been shown that the errors in the s-step version follow the same structure as the errors in the classical algorithm, but are amplified by a factor depending on the square of the condition number of the O(s)-dimensional Krylov bases computed in each outer loop. As the condition number of these s-step bases grows (in some cases very quickly) with s, this limits the s values that can be chosen and thus can limit the attainable performance. In this work, we show that if a select few computations in s-step Lanczos are performed in double the working precision, the error terms then depend only linearly on the conditioning of the s-step bases. This has the potential for drastically improving the numerical behavior of the algorithm with little impact on per-iteration performance. Our numerical experiments demonstrate the improved numerical behavior possible with the mixed precision approach, and also show that this improved behavior extends to mixed precision s-step CG. Here, we present preliminary performance results on NVIDIA V100 GPUs that show that the overhead of extra precision is minimal if one uses precisions implemented in hardware.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Windowed Fourier transform and cross-correlation algorithms for molecular tagging velocimetry

In this study, simulated and experimental molecular tagging velocimetry (MTV) images have been analyzed with a technique commonly used to process grid images on surfaces, the windowed Fourier transform with local spectrum analysis (WFT-LSA). A systematic synthetic image study of the modulation transfer function (MTF) and error tendencies of the WFT-LSA was performed and compared with a PIV-style cross-correlation algorithm to see if advanced strategies such as iterative image deformation can improve analysis of gridded images with high noise levels. Testing of single-pass algorithms showed that in typical MTV images, the WFT-LSA yields significantly lower bias errors than cross-correlation (CC) at displacements greater than 1 pixel but slightly higher random error at all displacements and image conditions. Analysis of the MTF shows that CC provided better resolution of spatial fluctuations than the WFT-LSA in many combinations of grid size and interrogation window. Tests of image deformation algorithms showed that the gap in performance between CC and WFT-LSA is maintained even as both methods improve. Additionally, WFT-LSA and CC methods are applied to real MTV experiments in high-speed gas jet flows. A preliminary analysis of phosphorescence lifetime provided by acetone vapor excited at 266 nm is used for assessing the required gas speeds for making MTV application feasible. The application of WFT-LSA to real MTV images demonstrates the ability of the algorithm to handle further real-world effects that could not be considered in the synthetic image analysis, like reduced signal-to-noise ratio and non-uniform intensity of the tagging grid across the image introduced by the actual laser beam energy distribution. With experimental images, CC is more accurate with shear flows but less robust to high noise levels than WFT-LSA, as predicted by the synthetic image analysis.

42 ENGINEERING↗
DOE OSTI2020

Distribution Development for the RDX Regional Model at Los Alamos National Laboratory - 20385

Representing uncertainty in model inputs often means finding a balance between uncertainty and physical reality. Developing wide distributions may seem conservative in principle, but this approach may lead to unrealistic model results. Characterizing the current state of knowledge of stochastic inputs presents many challenges, especially if the data available are limited or have limited relevance to the site. Relationships among these inputs may also be important to represent but are typically complex or difficult to define. Often special adjustments must be made to account for reduced credibility in particular data. If parameters are strongly related to one another, a correlation structure may be developed for input to the model. Other techniques such as regression models may be used to incorporate relationships between the information available and the desired parameters. The process of developing distributions must consider details of the model in terms of what the distribution is meant to represent. This paper uses the example of a probabilistic fate and transport model for hexahydro-1,3,5-trinitro-1,3,5-triazine (RDX) in the regional aquifer at Los Alamos National Laboratory (LANL). For many parameters, a single draw is applied to all space and time over which the model is run, for a single iteration. This simplification is often made for many reasons, and can often be beneficial, but also adds additional complexity in the distribution development process. Defining the distributional goals as they relate to the modeling process is an important step, which should take place prior to evaluation of the data. Usually, the distributions developed are meant to characterize the average value of the parameter over the spatial and temporal domain of the model. Distribution development requires consideration of many sources of information on the parameter where available, ideally from multiple references. Examples of different sources include data from different references but also from different conditions, measurement methods, or experimental types. Depending on these conditions and the reliability or relevance of particular references, different sources of data may each contribute valuable information but have varying relevance to the site. In these cases, weighting data unequally is a useful way to incorporate this information. As an example, aqueous dispersivity data are available for a variety of rock types. Only a few values are available for the desired rock type, and this is not enough to develop a distribution. Therefore, dispersivity values from other rock types are included in distribution development but are down-weighted such that the best data have the most influence on the distribution developed. In another case, K{sub d} distributions in the model are meant to represent a known composition of multiple soil types. Data from these materials are weighted accordingly to develop a distribution for the weighted average K{sub d} across soil types. Other cases include varying reliability of different sources, and weighting data according to the confidence in these sources. In some cases, input parameters are correlated with one another. An example is advective porosity, which is positively correlated with total porosity and must be less than total porosity. Paired data with both parameters must exist to discern the relationship between the two parameters and if it is necessary to build a correlation structure into the model. In general, correlated parameters may be represented in the model by a multivariate distribution, or perhaps more desirably, capturing the correlation within the developed distributions which may be treated as independent from one another. In the example of porosity, this can be done by transforming advective porosity into a proportion of total porosity which may be drawn independently from the distribution of total porosity. This paper explores how complex data and correlations can be incorporated to meet the distributional goals of the model, using the RDX regional model as a detailed example. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗
DOE OSTI2025

FLORA Equilibrium and Stability Code Archive

FLORA solves, in a 2-D domain (radial and axial dimensions) with a specified azimuthal Fourier mode, for the linearized stability of a long, thin, axisymmetric plasma equilibrium in an applied magnetic field. Before the stability equation is solved, FLORA solves a set of simple equations for pressure balance that specify the equilibrium magnetic field and plasma pressure in the long-thin limit given a simplified description of the magnetic coils. It uses an initial-value method for the linear stability problem in which an equilibrium is given an initial perturbation to its magnetic field, and the temporal behavior of the perturbation is followed. The perturbation has been Fourier expanded in the azimuthal coordinate; each azimuthal mode must be examined separately. The complex partial differential equation of motion for the perturbed radial displacement of the field lines is solved as a coupled system of two real p.d.e.'s and the solution consists of two parts, the real part and the imaginary part. The system is solved by bringing the coupling terms in each equation to the right side and using an iterative technique.

Cohen, Bruce [Lawrence Livermore National Laborato↗
DOE OSTI2021

Impact of graph structures for QAOA on MaxCut

The quantum approximate optimization algorithm (QAOA) is a promising method of solving combinatorial optimization problems using quantum computing. QAOA on the MaxCut problem has been studied extensively on graphs with specific structure; however, little is known about the general performance of the algorithm on arbitrary graphs. Here, we investigate how different graph characteristics correlate with QAOA performance at depths at most three on the MaxCut problem for all connected non-isomorphic graphs with at most eight vertices. Some good predictors of QAOA success relate to graph symmetries, odd cycles, and density. For example, on eight vertex graphs, the average probability for selecting an optimal solution for graphs that contain no odd cycles after three iterations of QAOA is 60.6% compared to 48.2% for those that do. The data generated from these studies are shared in a publicly accessible database to serve as a benchmark for QAOA calculations and experiments. Knowing the relationship between structure and performance can be used to identify classes of combinatorial problems that are likely to exhibit a quantum advantage.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Progress with applications of three-ion ICRF scenarios for fusion research: A review

The viability of magnetic confinement fusion as an energy source depends on achieving the high ion temperatures required for D-T fusion. Among the available techniques, plasma heating with waves in the ion cyclotron range of frequencies (ICRF) is a prominent method for bulk ion heating in fusion plasmas. Furthermore, a detailed understanding of the non-linear physics of alpha heating and the complex impact of MeV-range fast ions on plasma dynamics becomes progressively more important. This paper provides a comprehensive overview of recent developments with the three-ion ICRF scenarios on Alcator C-Mod, ASDEX Upgrade and JET tokamaks. The results demonstrate the flexibility of these novel scenarios for heating bulk ions in D-T ≈ 50%-50% plasmas and efficient generation of MeV-range fast ions in multi-ion species plasmas. In conclusion, several key results relevant for ITER and future fusion reactors are highlighted.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2020

A novel active optimization approach for rapid and efficient design space exploration using ensemble machine learning

In this work, a novel design optimization technique based on active learning, which involves dynamic exploration and exploitation of the design space of interest using an ensemble of machine learning algorithms, is presented. In this approach, a hybrid methodology incorporating an explorative weak learner (regularized basis function model) which fits high-level information about the response surface, and an exploitative strong learner (based on committee machine) that fits finer details around promising regions identified by the weak learner, is employed. For each design iteration, an aristocratic approach is used to select a set of nominees, where points that meet a threshold merit value as predicted by the weak learner are selected to be evaluated using expensive function evaluation. In addition to these points, the global optimum as predicted by the strong learner is also evaluated to enable rapid convergence to the actual global optimum once the most promising region has been identified by the optimizer. This methodology is first tested by applying it to the optimization of a two-dimensional multi-modal surface. The performance of the new active learning approach is compared with traditional global optimization methods, namely micro-genetic algorithm (pGA) and particle swarm optimization (PSO). It is demonstrated that the new optimizer is able to reach the global optimum much faster with a significantly fewer number of function evaluations. Subsequently, the new optimizer is also applied to a complex internal combustion (IC) engine combustion optimization case with nine control parameters related to fuel injection, initial thermodynamic conditions, and in-cylinder flow. It is again found that the new approach significantly lowers the number of function evaluations that are needed to reach the optimum design configuration (by up to 80%) when compared to particle swarm and genetic algorithm-based optimization techniques.

Machine learning, active learning, design optimiza↗
DOE OSTI2024

IDAES-PSE 2.6.0 Release

The Institute for the Design of Advanced Energy Systems (IDAES) Integrated Platform is a versatile computational environment offering extensive process systems engineering (PSE) capabilities for optimizing the design and operation of complex, interacting technologies and systems. IDAES enables users to efficiently search vast, complex design spaces to discover the lowest cost solutions while supporting the full process modeling lifecycle, from conceptual design to dynamic optimization and control. The extensible, open platform empowers users to create models of novel processes and rapidly develop custom analyses, workflows, and end-user applications. IDAES-PSE 2.6.0 Release Highlights Upcoming Changes IDAES will be switching to the new Pyomo solver interface in the next release. Whilst this will hopefully be a smooth transition for most users, there are a few important changes to be aware of. The new solver interface uses a different version of the IPOPT writer (“ipopt_v2”) and thus any custom configuration options you might have set for IPOPT will not carry over and will need to be reset. By default, the new Pyomo linear presolver will be activated with ipopt_v2. Whilst are working to identify any bugs in the presolver, it is possible that some edge cases will remain. IDAES will begin deploying a new set of scaling tools and APIs over the next few releases that make use of the new solver writers. The old scaling tools and APIs will remain for backward compatibility but will begin to be deprecated. New Models, Tools and Features New Intersphinx extension automatically linking Jupyter notebook examples to project documentation New end-to-end diagnostics example demonstrated on a real problem New complementarity formulation for VLE with cubic equations of state, backward compatibility for old formulation New solver interface with presolve (ipopt_v2) in support of upcoming changes to the initialization and APIs methods, with default set to ipopt to maintain backwards compatibility; this will deprecate once all examples have been updated New forecaster and parameterized bidder methods within grid integration library Updated surrogates API and examples to support Keras 3, with backwards compatibility for older formats such as TensorFlow SavedModel (TFSM) Updated costing base dictionary to include the 2023 cost year index value Updated ProcessBlock to include information on the constructing block class Updated Flowsheet Visualizer to allow visualize() method to return value and functions Bug Fixes Fixed bug in the Modular Property Framework that would cause errors when trying to use phase-based material balances with phase equilibria. Fixed bug in Modular Properties Framework that caused errors when initializing models with non-vapor-liquid phase equilibria. Fixed typos flagged by June update to crate-ci/typos and removed DMF-related exceptions Minor corrections of units of measurement handling in power plant waste/transport costing expressions, control volume material holdup expressions, and BTX property package parameters Fixed throwing >7500 numpy deprecation warnings by replacing scalar value assignment with element extraction and item iteration calls Testing and Robustness Migrated slow tests (>10s) to integration, impacting test coverage but also yielding a nearly 30% decrease in local test runtime Pinned pint to avoid issues with older supported Python versions Pinned codecov versions to avoid tokenless upload behavior with latest version Bumped extensions to version 3.4.2 to allow pointing to non-standard install location Deprecations and Removals Python 3.8 is no longer supported. The supported Python versions are 3.9 through 3.12 The Data Management Framework (DMF) is no longer supported. Importing idaes.core.dmf will cause a deprecation warning to be displayed until the next release The SOFC Keras surrogates have been removed. The current version of the SOFC surrogate model in the examples repository is a PySMO Kriging model.

AS↗
DOE OSTI2022

Quantifying the atomistic free-volume morphology of materials with graph theory

Here, we introduce a new computational methodology for the identification and characterization of free volume within/around atomistic configurations. This scheme employs a three-stage workflow, by which spheres are iteratively grown inside of voxels, and ultimately converted to planar graphs, which are then characterized via a graph-based order parameter. Our approach is computationally efficient, physically intuitive, and universally transferable to any material system. Validation of our methodology is performed on several sets of materials problems: (1) classification of unique free volumes in various crystal phases, (2) autonomous detection and classification of complex surface defects during epitaxial growth simulations, (3) characterization of free volume defects in metals/alloys, and (4) quantification of the spatio-temporal behavior of nano-scale free volume morphologies as a function of both temperature and free-volume size. Our method accurately identifies and characterizes unique free volumes over a multitude of systems and length scales, indicating its potential for future use in understanding the relationship between free volume morphology and material properties under both static and dynamic conditions.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Developing solid-surface plasma facing components for pilot plants and reactors with replenishable wall claddings and continuous surface conditioning. Part A: concepts and questions

Abstract It is estimated that pilot plants and reactors may experience rates of net erosion and deposition of solid plasma facing component (PFC) material of 10 3 –10 5 kg yr −1 . Even if the net erosion (wear) problem can be solved, the redeposition of so much material has the potential for major interference with operation, including disruptions due to so-called ‘unidentified flying objects (UFOs)’ and unsafe dust levels. The potential implications appear to be no less serious than for plasma contact with the divertor target: a dust explosion or a major UFO-disruption could be as damaging for an actively-cooled deuterium-tritium (DT) tokamak as target failure. It will therefore be necessary to manage material deposits to prevent their fouling operation. This situation appears to require a fundamental paradigm shift with regard to meeting the challenge of taming the plasma–material interface: it appears that any acceptable solid PFC material will in effect be flow-through , like liquid–metal PFCs, although at far lower mass flow rates. Solid PFC material will have to be treated as a consumable , like brake pads in cars. ITER will use high-Z (tungsten) armor on the divertor targets and low-Z (beryllium) on the main walls. The ARIES-AT reactor design calls for a similar arrangement, but with SiC cladding on the main walls. Non-metallic low-Z refractory materials such as ceramics (graphite, SiC, etc) used as in situ replenishable, relatively thin—of order mm—claddings on a substrate which is resistant to neutron damage could provide a potential solution for the main walls, while reducing the risk of degrading the confined plasma. Separately, wall conditioning has proven essential for achieving high performance. For DT devices, however, standard methods appear to be unworkable, but recently powder droppers injecting low-Z material ∼continuously into discharges have been quite effective and may be usable in DT devices as well. The resulting massive generation of low-Z debris, however, has the same potential to seriously disrupt operation as noted above. Powder droppers provide a unique opportunity to carry out controlled studies on the management of low-Z slag in all current tokamaks, independent of whether their protection tiles use low-Z or high-Z material.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2026

Electromagnetic Analysis of ITER Electron Cyclotron Emission Model

The ITER electron cyclotron emission (ECE) diagnostic system is located at diagnostics shielding module 2 (DSM2), equatorial port 9 (EP9), to measure electron temperature profile and electron temperature fluctuations, and also assess nonthermal electron distributions via the oblique view. Therefore, ECE has both radial and oblique views with two hot sources for calibration and two couples of mirrors for two different optical views. It shall receive that the large electromagnetic (EM) loads up to 100-MN/m3 force density due to the eddy currents on the copper mirrors generated by the short transient plasma disruption. The simplified EM analysis model using magnetic field ( B ) data and flux variations (dB/dt) method based on the worst case of plasma disruption, MD_DW_EXP16MS_CATIII, has been developed to calculate the force and moments for each ECE component and support structure for EP9 DSM2 Bay2 and Bay3. The B and dB/dt methods use the constant B and dB/dt assumed for each local bay space, but the real B and dB/dt are basically varied along the radial direction of each bay. The complex global model with the local ECE components and support structure using Maxwell transient has been developed to do EM analysis to compare the results of the B and dB/dt models. The volumetric EM force density of the whole ECE components and support structure can be used for the subsequent structural analysis. In conclusion, combined with the thermal and nuclear loads and seismic and initial loads, the ECE integration analysis can be finalized for the operation case.

Fang, J. [Princeton Plasma Physics Laboratory (PPP↗
DOE OSTI2023

Revisiting Temporal Blocking Stencil Optimizations

Iterative stencils are used widely across the spectrum of High Performance Computing (HPC) applications. Many efforts have been put into optimizing stencil GPU kernels, given the prevalence of GPU-accelerated supercomputers. To improve the data locality, temporal blocking is an optimization that combines a batch of time steps to process them together. Under the observation that GPUs are evolving to resemble CPUs in some aspects, we revisit temporal blocking optimizations for GPUs. We explore how temporal blocking schemes can be adapted to the new features in the recent Nvidia GPUs, including large scratchpad memory, hardware prefetching, and device-wide synchronization. We propose a novel temporal blocking method, EBISU, which champions low device occupancy to drive aggressive deep temporal blocking on large tiles that are executed tile-by-tile. We compare EBISU with state-of-the-art temporal blocking libraries: STENCILGEN and AN5D. We also compare with state-of-the-art stencil auto-tuning tools that are equipped with temporal blocking optimizations: ARTEMIS and DRSTENCIL. Over a wide range of stencil benchmarks, EBISU achieves speedups up to 2.53x and a geometric mean speedup of 1.49x over the best state-of-the-art performance in each stencil benchmark.

Zhang, Lingqi↗
DOE OSTI2023

AMG Preconditioners based on parallel hybrid coarsening and multi-objective graph matching

We describe preliminary results from a multi-objective graph matching algorithm, in the coarsening step of an aggregation-based Algebraic MultiGrid (AMG) preconditioner, for solving large and sparse linear systems of equations on high-end parallel computers. We have two objectives. First, we wish to improve the convergence behavior of the AMG method when applied to highly anisotropic problems. Second, we wish to extend the parallel package \texttt{PSCToolkit} to exploit multi-threaded parallelism at the node level on multi-core processors. Our matching proposal balances the need to simultaneously compute high weights and large cardinalities by a new formulation of the weighted matching problem combining both these objectives using a parameter $\lambda$. We compute the matching by a parallel $2/3-\varepsilon$-approximation algorithm for maximum weight matchings. Results with the new matching algorithm show that for a suitable choice of the parameter $\lambda$ we compute effective preconditioners in the presence of anisotropy, i.e., smaller solve times, setup times, iterations counts, and operator complexity.

D'Ambra, Pasqua↗
DOE OSTI2022

Machine Learning-driven Molecular Design for Therapeutic Discovery

The ongoing novel coronavirus pandemic (COVID-19) has highlighted the need for new therapeutics to counter the threat of emerging viral pathogens. The main proteases are a promising target for developing antiviral inhibitors. In this work, we utilized a novel combination of artificial intelligence-driven iterative design of covalent inhibitor candidates, physics-based computational modeling of protein-inhibitor interactions, and “All in One” Native MS biophysical assay screening and characterization of therapeutic candidates. With our existing expertise in hit generation using a particular scaffold as a starting point, we first generated tens of thousands of compounds that preserve the key scaffold. In order to optimize the candidates, we calculated about 136 descriptors consisting of 2D and 3D features for molecules targeting the SARS-CoV-2 Main protease (Mpro). These compounds were initially filtered according to properties and further sorted by predicted binding affinity using our automated docking modeling and machine learning methods. We tested a handful of candidates and identified two as inhibitors of Mpro with micromolar affinities.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2024

Distributed optimization for multi-commodity urban traffic control

A distributed method for concurrent traffic signal and routing control of traffic networks is proposed. The method is based on the multi-commodity store-and-forward model, in which the destinations are the commodities. The system benefits from the communication between vehicles and infrastructure, providing optimal signal timings to intersections and routes to vehicles on a link-by-link basis. Using the augmented Lagrangian to model the constraints into the objective, the baseline centralized problem is decomposed into a set of objective-coupled subproblems, one for each intersection, enabling the solution to be computed by a distributed- gradient projection algorithm. Further, the intersection agents only need to communicate and coordinate with neighboring intersections to ensure convergence to the optimal solution while tolerating suboptimal iterations that offer more flexibility, unlike other distributed approaches. Through microsimulation, we demonstrate the effectiveness of the proposed algorithm in traffic networks with time-varying demand. Computational analysis shows that the distributed problem is suitable for real-time applications. A robustness analysis show that the distributed formulation enables a graceful degradation of the system in case of failure.

Augmented Lagrangian↗
DOE OSTI2020

Deep learning to generate in silico chemical property libraries and candidate molecules for small molecule identification in complex samples

Comprehensive and unambiguous identification of small molecules in complex samples will revolutionize our understanding of the role of metabolites in biological systems. Existing and emerging technologies have enabled measurement of chemical properties of molecules in complex mixtures and, in concert, are sensitive enough to resolve even stereoisomers. Despite these experimental advances, small molecule identification is inhibited by (i) chemical reference libraries (e.g. mass spectra, collision cross section, and other measurable property libraries) representing <1% of known molecules, limiting the number of possible identifications, and (ii) the lack of a method to generate candidate matches directly from experimental features (i.e. without a library). To this end, we developed a variational autoencoder (VAE) to learn a continuous numerical, or latent, representation of molecular structure to expand reference libraries for small molecule identification. We extended the VAE to include a chemical property decoder, trained as a multitask network, in order to shape the latent representation such that it assembles according to desired chemical properties. The approach is unique in its application to metabolomics and small molecule identification, with its focus on properties that can be obtained from experimental measurements (m/z, CCS) paired with its training paradigm, which involved a cascade of transfer learning iterations. First, molecular representation is learned from a large dataset of structures with m/z labels. Next, in silico property values are used to continue training, as experimental property data is limited. Finally, the network is further refined by being trained with the experimental data. This allows the network to learn as much as possible at each stage, enabling success with progressively smaller datasets without overfitting. Once trained, the network can be used to predict chemical properties directly from structure, as well as generate candidate structures with desired chemical properties. Our approach is orders of magnitude faster than first-principles simulation for CCS property prediction. Additionally, the ability to generate novel molecules along manifolds, defined by chemical property analogues, positions DarkChem as highly useful in a number of application areas, including metabolomics and small molecule identification, drug discovery and design, chemical forensics, and beyond.

Colby, Sean M.↗
DOE OSTI2025

Benders Decomposition Using Graph Modeling and Multi-Parametric Programming

Benders decomposition is a widely used method for solving large and structured optimization problems, but its performance is affected by the repeated solution of subproblems. We propose a flexible and modular algorithmic framework for accelerating Benders decomposition. Specifically, we express the problem structure by using a graph-theoretic modeling abstraction in which nodes represent optimization subproblems and edges represent connectivity between subproblems. A key innovation of our approach is that we embed multiparametric programming (mp) surrogates for node subproblems, which maps the exact analytical map of the subproblem solution space. The use of mp surrogates allows us to replace subproblem solves with fast look-ups and function evaluations for primal and dual variables during the iterative Benders process. We formally show the equivalence between classical Benders cuts and those derived from the mp solution. We implement our framework in the open-source PlasmoBenders.jl software package. To demonstrate the capabilities of the proposed framework, we apply it to a two-stage stochastic programming problem, which aims to make optimal capacity expansion decisions under market uncertainty. We evaluate both single-cut and multicut variants of Benders decomposition and show that the use of mp surrogates achieves substantial speedups in subproblem solve time, while preserving the convergence guarantees of Benders decomposition. We highlight advantages in solution analysis and interpretability that is enabled by mp critical region tracking; specifically, we show that these reveal how decisions evolve geometrically across the Benders search. Our results aim to demonstrate that combining surrogate modeling with graph modeling offers a promising and extensible foundation for structure-exploiting decomposition. In addition, by decomposing the problem into more tractable subproblems, the proposed approach also aims to overcome scalability issues of mp. Finally, the use of mp surrogates provides a unifying and modular optimization framework that enables the representation of heterogeneous node subproblems as modeling objects with a homogeneous structure.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗