Search NASA⌕ Search

SEARCH · Search NASA

Results for “Dynamic clustering algorithm”

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 217 records · Page 12

Elastic distributed training with fast convergence and efficient resource utilization

Distributed learning is now routinely conducted on cloud as well as dedicated clusters. Training with elastic resources brings new challenges and design choices. Prior studies focus on runtime performance and assume a static algorithmic behavior. In this work, by analyzing the impact of of resource scaling on convergence, we introduce schedules for synchronous stochastic gradient descent that proactively adapt the number of learners to reduce training time and improve convergence. Our approach no longer assumes a constant number of processors throughout training. In our experiment, distributed stochastic gradient descent with dynamic schedules and reduction momentum achieves better convergence and significant speedups over prior static ones. Numerous distributed training jobs running on cloud may benefit from our approach.

Cong, Guojing↗

Protonation Dynamics of Confined Ethanol–Water Mixtures in H-ZSM-5 from Machine Learning-Driven Metadynamics

Zeolites are indispensable heterogeneous catalysts in industrial chemical processes, valued for their strong Brønsted acidity, well-defined microporous frameworks, and tunable pore structures. Their catalytic activity arises primarily from Brønsted acid sites (BAS), typically present as bridging hydroxyl groups (Si–OH–Al). Under aqueous reaction conditions, these protons interact dynamically with water and alcohol molecules, leading to complex solvation and protonation behavior within confined pores. In this study, we investigate the protonation equilibrium occurring between ethanol and water at the BAS of acidic zeolites under varying hydration levels, i.e., C2H5OH–(H2O)n, n=1–4. Local structure was analyzed through an adaptive-learning global optimization algorithm, while enhanced sampling molecular dynamics simulations with Well-Tempered Metadynamics (WMetaD) and machine learning interatomic potentials (MLPs) provide free-energy surfaces (FES) at variable hydration levels. The results reveal a strong dependence of proton localization on the degree of hydration. At low hydration (1 water molecule), the proton resides predominantly on ethanol; with 2 water molecules, it shifts toward water, and at higher hydration (3 or more water molecules), it becomes extensively delocalized over the water cluster. These findings underscore the critical role of solvation in modulating acid site behavior and suggest that a minimum of three water molecules is necessary to fully stabilize the proton on water within the zeolite framework. This solvation threshold has significant implications for catalytic processes, particularly in biomass conversion reactions where alcohol protonation is a key step in dehydration mechanisms.

machine learning↗

Direct diabatization and analytic representation of coupled potential energy surfaces and couplings for the reactive quenching of the excited 2 Σ + state of OH by molecular hydrogen

In this work, we have employed extended multiconfiguration quasidegenerate perturbation theory, fourfold-way diabatic molecular orbitals, and configurational uniformity to develop a global three-state diabatic representation of the potential energy surfaces and their couplings for the electronically nonadiabatic reaction OH* + H 2 → H 2 O + H, where * denotes electronic excitation to the A 2 Σ + state. To achieve sign consistency of the computed diabatic couplings, we developed a graphics processing unit-accelerated algorithm called the cluster-growing algorithm. Having obtained consistent signs of the diabatic couplings, we fit the diabatic matrix elements (which consist of the diabatic potentials and the diabatic couplings) to analytic representations. Adiabatic potential energy surfaces are generated by diagonalizing the 3 × 3 diabatic potential energy matrix. The comparisons between the fitted and computed diabatic matrix elements and between the originally computed adiabatic potential energy surfaces and those generated from the fits indicate that the current fit is accurate enough for dynamical studies, and it may be used for quantal or semiclassical dynamics calculations.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

NWChem: scalable parallel computational chemistry

NWChem is a general purpose computational chemistry code specifically designed to run on distributed memory parallel computers. The core functionality of the code focuses on molecular dynamics, Hartree-Fock and density functional theory methods for both plane-wave basis sets as well as Gaussian basis sets, tensor contraction engine based coupled cluster capabilities and combined quantum mechanics/molecular mechanics descriptions. It was realized from the beginning that scalable implementations of these methods required a programming paradigm inherently different from what message passing approaches could offer. In response a global address space library, the Global Array Toolkit, was developed. The programming model it offers is based on using predominantly one-sided communication. This model underpins most of the functionality in NWChem and the power of it is exemplified by the fact that the code scales to tens of thousands of processors. In this paper the core capabilities of NWChem are described as well as their implementation to achieve an efficient computational chemistry code with high parallel scalability. NWChem is a modern, open source, computational chemistry code1 specifically designed for large scale parallel applications2. To meet the challenges of developing efficient, scalable and portable programs of this nature a particular code design was adopted. This code design involved two main features. First of all, the code is build up in a modular fashion so that a large variety of functionality can be integrated easily. Secondly, to facilitate writing complex parallel algorithms the Global Array toolkit was developed. This toolkit allows one to write parallel applications in a shared memory like approach, but offers additional mechanisms to exploit data locality to lower communication overheads. This framework has proven to be very successful in computational chemistry but is applicable to any engineering domain. Within the context created by the features above NWChem has grown into a general purpose computational chemistry code that supports a wide variety of energy expressions and capabilities to calculate properties based there upon. The main energy expressions are classical mechanics force fields, Hartree-Fock and DFT both for finite systems and condensed phase systems, coupled cluster, as well as QM/MM. For most energy expressions single point calculations, geometry optimizations, excited states, and other properties are available. Below we briefly discuss each of the main energy expressions and the critical points involved in scalable implementations thereof.

99 GENERAL AND MISCELLANEOUS↗

A solution algorithm for fluid–particle flows across all flow regimes

Many fluid–particle flows occurring in nature and in technological applications exhibit large variations in the local particle volume fraction. For example, in circulating fluidized beds there are regions where the particles are closepacked as well as very dilute regions where particle–particle collisions are rare. Thus, in order to simulate such fluid–particle systems, it is necessary to design a flow solver that can accurately treat all flow regimes occurring simultaneously in the same flow domain. In this work, a solution algorithm is proposed for this purpose. The algorithm is based on splitting the free-transport flux solver dynamically and locally in the flow. In close-packed to moderately dense regions, a hydrodynamic solver is employed, while in dilute to very dilute regions a kinetic-based finite-volume solver is used in conjunction with quadrature-based moment methods. To illustrate the accuracy and robustness of the proposed solution algorithm, it is implemented in OpenFOAM for particle velocity moments up to second order, and applied to simulate gravity-driven, gas–particle flows exhibiting cluster-induced turbulence. By varying the average particle volume fraction in the flow domain, it is demonstrated that the flow solver can handle seamlessly all flow regimes present in fluid–particle flows.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Mapping Strain-rate Dependent Dislocation-Defect Interactions by Atomistic Simulations

Probing the mechanisms of defect-defect interactions at strain rates lower than 106 s-1 is an unresolved challenge to date to molecular dynamics (MD) techniques. Here we propose a novel atomistic approach based on transition state theory and the concept of a strain-dependent effective activation barrier that is capable of simulating the kinetics of dislocation-defect interactions at virtually any strain rate, exemplified within 10-7 to 107 s-1. We apply this approach to the problem of an edge dislocation colliding with a cluster of self-interstitial atoms (SIA) under shear deformation. Using an activation-relaxation algorithm (1), we uncover a unique strain-rate dependent trigger mechanism that allows the SIA cluster to be absorbed during the process leading to dislocation climb. Guided by this finding, we determine the activation barrier of the trigger mechanism as a function of shear strain, and use that in a coarse-graining rate equation formulation for constructing a mechanism map in the phase space of strain-rate and temperature. Our predictions of a crossover from a defect recovery at the low strain rate regime to defect absorption behavior in the high strain-rate regime are validated against our own independent, direct MD simulations at 105 to 107 s-1. Implications of the present approach for probing molecular-level mechanisms in strain-rate regimes previously considered inaccessible to atomistic simulations are discussed.

activation mecanism↗

Shape effects on the local dynamics of suspensions of spheroidal particles

The effect of shape on the dynamics of suspensions of non-spherical heavy particles is examined by fully resolved numerical simulations of oblate and prolate spheroids, as well as spheres, for a density ratio of ten, volume fractions ranging from 0.5% to 5%, and Reynolds numbers between 20 and 30. The dynamics is determined both by the interactions of the particles with the fluid as well as by collisions, with the number and importance of collisions increasing with volume fractions. A single isolated oblate or prolate spheroid falling under gravity generally falls broadside on, for the governing parameters examined here, and at low-volume fractions, the majority of particles in a suspension fall that way. At higher-volume fractions, the orientation is more random. The slip velocity decreases as the volume fraction increases for all shapes, as expected, but the effect of the shape is much less than seen for a single particle. Furthermore, this seems to be due to two effects. For all volume fractions, the most deformed particles cluster more than spheres and less deformed particles, which increases their slip velocity. As the concentration increases, the increased particle interactions also causes more particles to fall short side-on, which reduces the frontal area and the resulting drag, increasing the slip velocity. This second effect is, of course, absent for spherical particles.

42 ENGINEERING↗

Autonomous Experiments in Scanning Probe Microscopy and Spectroscopy: Choosing Where to Explore Polarization Dynamics in Ferroelectrics

Polarization dynamics in ferroelectric materials are explored via automated experiment in piezoresponse force microscopy/spectroscopy (PFM/S). A Bayesian optimization (BO) framework for imaging is developed, and its performance for a variety of acquisition and pathfinding functions is explored using previously acquired data. The optimized algorithm is then deployed on an operational scanning probe microscope (SPM) for finding areas of large electromechanical response in a thin film of PbTiO 3 , with results showing that, with just 20% of the area sampled, most high-response clusters were captured. Furthermore, this approach can allow performing more complex spectroscopies in SPM that were previously not possible due to time constraints and sample stability. Improvements to the framework to enable the incorporation of more prior information and improve efficiency further are modeled and discussed.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Materials Learning Algorithms (MALA): Scalable machine learning for electronic structure calculations in large-scale atomistic simulations

We present the Materials Learning Algorithms (MALA) package, a scalable machine learning framework designed to accelerate density functional theory (DFT) calculations suitable for large-scale atomistic simulations. Using local descriptors of the atomic environment, MALA models efficiently predict key electronic observables, including local density of states, electronic density, density of states, and total energy. The package integrates data sampling, model training and scalable inference into a unified library, while ensuring compatibility with standard DFT and molecular dynamics codes. We demonstrate MALA's capabilities with examples including boron clusters, aluminum across its solid-liquid phase boundary, and predicting the electronic structure of a stacking fault in a large beryllium slab. Scaling analyses reveal MALA's computational efficiency and identify bottlenecks for future optimization. With its ability to model electronic structures at scales far beyond standard DFT, MALA is well suited for modeling complex material systems, making it a versatile tool for advanced materials research.

Density functional theory↗

Dynamics-Based Peptide–MHC Binding Optimization by a Convolutional Variational Autoencoder: A Use-Case Model for CASTELO

An unsolved challenge in the development of antigen-specific immunotherapies is determining the optimal antigens to target. Comprehension of antigen–major histocompatibility complex (MHC) binding is paramount toward achieving this goal. Here, we apply CASTELO, a combined machine learning-molecular dynamics (ML-MD) approach, to identify per-residue antigen binding contributions and then design novel antigens of increased MHC-II binding affinity for a type 1 diabetes-implicated system. We build upon a small-molecule lead optimization algorithm by training a convolutional variational autoencoder (CVAE) on MD trajectories of 48 different systems across four antigens and four HLA serotypes. We develop several new machine learning metrics including a structure-based anchor residue classification model as well as cluster comparison scores. ML-MD predictions agree well with experimental binding results and free energy perturbation-predicted binding affinities. Moreover, ML-MD metrics are independent of traditional MD stability metrics such as contact area and root-mean-square fluctuations (RMSF), which do not reflect binding affinity data. Finally, our work supports the role of structure-based deep learning techniques in antigen-specific immunotherapy design.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Small-amplitude Compressible Magnetohydrodynamic Turbulence Modulated by Collisionless Damping in Earth’s Magnetosheath: Observation Matches Theory

Plasma turbulence is a ubiquitous dynamical process that transfers energy across many spatial and temporal scales and affects energetic particle transport. Recent advances in the understanding of compressible magnetohydrodynamic (MHD) turbulence demonstrate the important role of damping in shaping energy distributions on small scales, yet its observational evidence is still lacking. This study provides the first observational evidence of substantial collisionless damping (CD) modulation on the small-amplitude compressible MHD turbulence cascade in Earth’s magnetosheath using four Cluster spacecraft. Based on an improved compressible MHD decomposition algorithm, turbulence is decomposed into three eigenmodes: incompressible Alfvén modes and compressible slow and fast (magnetosonic) modes. Our observations demonstrate that CD enhances the anisotropy of compressible MHD modes because CD has a strong dependence on wave propagation angle. The wavenumber distributions of slow modes are mainly stretched perpendicular to the background magnetic field (B 0 ) and weakly modulated by CD. In contrast, fast modes are subjected to a more significant CD modulation. Fast modes exhibit a weak, scale-independent anisotropy above the CD truncation scale. Below the CD truncation scale, the anisotropy of fast modes enhances as wavenumbers increase. As a result, fast-mode fractions in the total energy of compressible modes decrease with the increase of perpendicular wavenumber (to B 0 ) or wave propagation angle. Our findings reveal how the turbulence cascade is shaped by CD and its consequences for anisotropies in the space environment.

97 MATHEMATICS AND COMPUTING↗

A universal state-selective approach to multireference coupled-cluster non-iterative corrections

A new form of the asymmetric energy functional for the multireference coupled cluster (MRCC) theories is discussed from the point of view of an energy expansion in quasidegenerate situation. The resulting expansion for the exact electronic energy can be used to define the non-iterative corrections to approximate MRCC approaches. In particular, we show that in the proposed framework the essential part of dynamic correlation can be encapsulated in the so-called correlation Hamiltonian, which in analogy to the effective Hamiltonian, is defined in the model space (M0). When the trail wavefunction (wavefunctions) is determined without invoking a specific form of the MRCC sufficiency conditions, the ensuing correction can be universally applied to any type of the approximate MRCC method employing the same excitation manifold. We also discuss several strategies of normalizing trial wavefunctions, which significantly simplify the algebraic structure of the correction. The analogies with other MRCC triples corrections to MRCC theories with singles and doubles are outlined. Possible parallel algorithms for the non-iterative corrections are alsodiscussed.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

A comparative analysis of clustering algorithms: O{sub 2} migration in truncated hemoglobin I from transition networks

The ligand migration network for O{sub 2}–diffusion in truncated Hemoglobin N is analyzed based on three different clustering schemes. For coordinate-based clustering, the conventional k–means and the kinetics-based Markov Clustering (MCL) methods are employed, whereas the locally scaled diffusion map (LSDMap) method is a collective-variable-based approach. It is found that all three methods agree well in their geometrical definition of the most important docking site, and all experimentally known docking sites are recovered by all three methods. Also, for most of the states, their population coincides quite favourably, whereas the kinetics of and between the states differs. One of the major differences between k–means and MCL clustering on the one hand and LSDMap on the other is that the latter finds one large primary cluster containing the Xe1a, IS1, and ENT states. This is related to the fact that the motion within the state occurs on similar time scales, whereas structurally the state is found to be quite diverse. In agreement with previous explicit atomistic simulations, the Xe3 pocket is found to be a highly dynamical site which points to its potential role as a hub in the network. This is also highlighted in the fact that LSDMap cannot identify this state. First passage time distributions from MCL clusterings using a one- (ligand-position) and two-dimensional (ligand-position and protein-structure) descriptor suggest that ligand- and protein-motions are coupled. The benefits and drawbacks of the three methods are discussed in a comparative fashion and highlight that depending on the questions at hand the best-performing method for a particular data set may differ.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Global teleconnections influencing large-scale drought in the United States using SVDI

Understanding recent large-scale drought patterns and the mechanisms producing extreme drought events is vital for future drought forecasts and understanding future drought risks. Increasingly, vapor pressure deficit (VPD) has been used as an important measure of evaporative demand and proxy for drought detection. In this study, VPD is used to calculate the new Standardized VPD Drought Index (SVDI) with NASA North American Land Data Assimilation System (NLDAS) data. Previous studies have shown that SVDI accurately identifies the timing and magnitude short-term droughts in the United States (U.S). In the present study, SVDI is now used to identify large-scale drought patterns between 1980 and 2021 and drought variability driven by selected global teleconnections originating in the Pacific and Atlantic Oceans. Spatial drought characteristics were extracted from SVDI using empirical orthogonal function (EOF) analysis. Then a k-means clustering algorithm was applied to both EOF principal components and primary teleconnections, including the El Nino-Southern Oscillation (ENSO) and Pacific Decadal Oscillation (PDO) to identify drought events driven by the Pacific Ocean. Results show that the SVDI is useful in evaluating large-scale drought variability in the U.S. related to global teleconnections, and that mechanisms influencing summer drought patterns in the Western and Southwestern U.S. are driven by a tropical-extratropical interactions originating in the equatorial Pacific Ocean related to ENSO dynamics with interdecadal variability modulated by PDO. The large-scale droughts in the Central and Southern U.S., like those in 2011 and 2012, on the other hand, are driven by the North Pacific Ocean warm pool during a strong negative PDO, which subsequently influenced variability in the Bermuda-Azores High in the Atlantic Ocean. In summer 2011, the Bermuda-Azores High weakened, reducing the onshore winds and moisture transport along the eastern Gulf of Mexico and contributing to ongoing drought in the region. The Northern Pacific and Atlantic Ocean sea surface temperatures (SSTs) have increased between 1980 and 2021. In conclusion, as SSTs continue to rise in the Northern Pacific Ocean, one consequence of the coupled North Pacific warm pool and atmospheric dynamics, is to increase summer drought variability over a large region in the southern and midwestern U.S. under global warming.

54 ENVIRONMENTAL SCIENCES↗

Intelligent Experiments through Real-Time AI: Fast Data Processing and Autonomous Detector Control for High-Energy Nuclear Experiments

The aim of this project is to develop software and hardware for fast real-time data processing and autonomous detector control and calibration for the sPHENIX and the future EIC experiments. Below summarizes Georgia Tech team efforts in the past year: 1. We developed a real-time clustering algorithm and FPGA-based pipeline architecture for processing fired pixel data from ALPIDE sensors in sPHENIX experiments. Our Columnar Clustering Co-Design introduces a hardware-aware, stream-friendly approach that segments pixel data by column pairs using a Column Pair Clustering (CPC) strategy, followed by Cluster Stitching to merge adjacent subclusters. Implemented in Vitis HLS, the pipeline comprises five stages—read-in, subclustering, stitching, analysis, and write-out—connected by tagged HLS streams with custom end-of-event signaling for robust synchronization. We designed a pipelined dataflow model optimized for throughput, low latency, and minimal buffering, enabling scalable clustering across events of arbitrary size. Our system maintains spatial precision via center-of-mass and shape key extraction and efficiently handles edge cases such as fragmented or nested clusters. Compared against DBSCAN in both software and hardware, our approach demonstrates competitive performance under FPGA constraints. 2. We also conducted a comprehensive algorithm-to-hardware co-design of connected component analysis tailored for sPHENIX experiments, focusing on real-time, low-latency processing using FPGAs and High-Level Synthesis (HLS). Starting from a Python-based particle tracking pipeline, the team translated the core logic—graph traversal via DFS and Union-Find—into an HLS-compatible C++ model, replacing dynamic memory and recursion with static arrays and pipelined control flow. The final design includes a fully streamed and dataflow-compatible Union-Find kernel optimized across five iterations, incorporating loop pipelining, array partitioning, AXI/FIFO interface tuning, and function flattening. Experimental results show up to 14.8× speedup over the CPU baseline, reducing per-graph latency to 1.58 μs and demonstrating strong resource efficiency with only ~7k LUTs and zero BRAM usage. The design maintains functional correctness against the Python reference using a Python-based C-simulation framework and Mean Squared Error metrics. This work validates the potential of HLS-driven FPGA designs for edge-level HEP data acquisition, laying a scalable foundation for future integration with real-time detector pipelines and multi-graph processing systems.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

An Orthogonal Recursive Bisection (ORB) Based Time Advancement Algorithm for CFD-DEM Solvers

The time integration of the granular phase in coupled computational fluid dynamics (CFD) – discrete element method (DEM) simulations presents a unique computational challenge brought about by the large variations in particle collisional time scales. Particles in the dilute regions of the computational domain can be advanced with large time steps while dense regions require much smaller time increments. However, the time step size in most solvers is globally set as the limit for accuracy and stability imposed by the collisions and is typically orders of magnitude less than that required away from collisions. This work addresses this precise issue and provides a strategy to avoid the use of a global conservative small time step size for the entire set of particles.A novel time stepping algorithm for CFD-DEM solvers using a partitioning approach using orthogonal recursive bisection (ORB) that allows for variable time steps among particles is described and its computational performance is compared against baseline explicit methods, typically used in several CFD-DEM solvers. ORB has advantages of being relatively quick and easy to update incrementally and has the required heuristic behavior (i.e., it will split the region in half with a cluster on each side) when groups of particles are well separated (clustered). The algorithm presented in this work uses a local time stepping approach to resolve collisional time scales for subsets of particles that are present at the leaves of the ORB, thereby resulting in substantial reduction of computational cost. The parallel implementation of this method where a ``knapsack” algorithm is used in tandem with ORB for effective load-balancing is also presented, where a best possible partitioning is obtained based on number of particles and local time-stepping costs. The algorithm is tested against benchmark problems with varying particle distributions that include fluidized bed and riser flow scenarios. Preliminary results indicate that the approach is 2-3X faster than traditional explicit methods for problems that involve both dense and dilute regions, while maintaining the same level of accuracy.

adaptive timestepping↗

Variational approaches to constructing the many-body nuclear ground state for quantum computing

Here, we explore the preparation of specific nuclear states on gate-based quantum hardware using variational algorithms. Large-scale classical diagonalizations of the nuclear shell model have reached sizes of 10 9 –10 10 basis states but are still severely limited by computational resources. Quantum computing can, in principle, solve such systems exactly with exponentially fewer resources than classical computing. Exact solutions for large systems require many qubits and large gate depth, but variational approaches can effectively limit the required gate depth. We use the unitary coupled cluster approach to construct approximations of the ground-state vectors, later to be used in dynamics calculations. The testing ground is the phenomenological shell model space, which allows us to mimic the complexity of the internucleon interactions. We find that often one needs to minimize over a large number of parameters, using a large number of entanglements that makes the application on existing hardware challenging. Prospects for rapid improvements with more capable hardware are, however, very encouraging.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Efficiently sampling conformations and pathways using the concurrent adaptive sampling (CAS) algorithm

Molecular dynamics (MD) simulations are useful in obtaining thermodynamic and kinetic properties of bio-molecules but are limited by the timescale barrier, i.e., we may be unable to efficiently obtain properties because we need to run microseconds or longer simulations using femtoseconds time steps. While there are several existing methods to overcome this timescale barrier and efficiently sample thermodynamic and/or kinetic properties, problems remain in regard to being able to sample un- known systems, deal with high-dimensional space of collective variables, and focus the computational effort on slow timescales. Hence, in this work, a new sampling method, called the “Concurrent Adaptive Sampling (CAS) algorithm,” has been developed to tackle these three issues and efficiently obtain conformations and pathways. The method is not constrained to use only one or two collective variables, unlike most reaction coordinate-dependent methods. Instead, it can use a large number of collective vari- ables and uses macrostates (a partition of the collective variable space) to enhance the sampling. The exploration is done by running a large number of short simula- tions, and a clustering technique is used to accelerate the sampling. In this paper, we introduce the new methodology and show results from two-dimensional models and bio-molecules, such as penta-alanine and triazine polymer

59 BASIC BIOLOGICAL SCIENCES↗