Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “synchronous machine”

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 307 records · Page 17
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Virtual sensing-enabled digital twin framework for real-time monitoring of nuclear systems leveraging deep neural operators

Abstract Real-time monitoring is a foundation of nuclear digital twin technology, crucial for detecting material degradation and maintaining nuclear system integrity. Traditional physical sensor systems face limitations, particularly in measuring critical parameters in hard-to-reach or harsh environments, often resulting in incomplete data coverage. Machine learning-driven virtual sensors offer a transformative solution by complementing physical sensors in monitoring critical degradation indicators. This paper introduces the use of Deep Operator Networks (DeepONet) to predict key thermal-hydraulic parameters in the hot leg of pressurized water reactor. DeepONet acts as a virtual sensor, mapping operational inputs to spatially distributed system behaviors without requiring frequent retraining. Our results show that DeepONet achieves low mean squared and Relative L2 error, making predictions 1400 times faster than traditional CFD simulations . These characteristics enable DeepONet to function as a real-time virtual sensor, synchronizing with the physical system to track degradation conditions and provide insights within the digital twin framework for nuclear systems.

Hossain, Raisa↗
DOE OSTI2026

Precision Polishing of Spheres Via In-Situ Process Monitoring and Machine-Learning-Based Optimization

In inertial confinement fusion (ICF) experiments seeking output gains of unity and beyond, the quality of the ablator capsule is paramount for minimizing hydrodynamic mix that quenches the central hot spot. Defects in the form of foreign particles or missing mass on the surface and within the wall of the capsule are primary offenders. High density carbon capsules made for ICF experiments on the National Ignition Facility (NIF) are precision polished to achieve the surface smoothness in the order of a few nm as well as to minimize isolated defects in the form of pits. Given the critical role of this process, we are developing smart manufacturing techniques with goal of elevating the efficiency of this process. Our approach is to use MEMS-based sensors to capture the fine vibrational signals generated during the polishing process and combine it with synchronized visual feedback as needed. Beyond using these sensors for process monitoring, we use specific deep learning methods to analyze the data and extract correlations with both the process parameters and the final performance of the polishing run. Here, we describe the multiple fronts that we have explored in this regard and the results we have gotten so far. This approach promises to have the potential to ultimately provide real-time feedback that can be used for ensuring the progress of the run as well as a means for faster optimization.

36 MATERIALS SCIENCE↗
NASA NTRS1982

Initial utility experience with cluster of three Mod-2 wind turbine systems

This paper describes the initial utility experiences of operating three MOD-2s during the Engineering Acceptance Testing. Electrical quantities of bus voltage, phase currents and power are initially being recorded to evaluate impacts to customers on the 69-kV subtransmission line during synchronization and operation of one or more WTSs. To date, effects on the system have been essentially undetectable. Measurements of television signal strengths were taken at an existing television remote pickup and relay station at the WTS site. Potential TV signal interference problems from the WTSs have been avoided by replacing the remote pickups with microwave repeater links for the four TV channels received from Portland, Oregon. Preliminary measurements of audible and sub-audible noise levels indicate that the upwind rotor, tubular tower design of the MOD-2 does not have the pulsing high intensity infrasound problems experienced by the MOD-1 machine at Boone, North Carolina.

Seely, D. B.↗
DOE OSTI2024

Precision Polishing of Ablator Capsules via in situ Process Monitoring and Machine Learning–Based Optimization

In inertial confinement fusion (ICF) experiments seeking output gains of unity and beyond, the quality of the ablator capsule is paramount for minimizing the hydrodynamic mix that quenches the central hot spot. Defects in the form of foreign particles or missing mass on the surface and within the wall of the capsule are primary offenders. High-density carbon capsules made for ICF experiments at the National Ignition Facility are precision polished to achieve surface smoothness on the order of a few nanometers as well as to minimize isolated defects in the form of pits. Given the critical role of this process, we are developing smart manufacturing techniques with the goal of elevating the efficiency of this process. Our approach is to use MEMS (micro-electromechanical systems)–based sensors to capture the fine vibration signals generated during the polishing process and combine them with synchronized visual feedback as needed. Beyond using these sensors for process monitoring, we use specific deep learning methods to analyze the data and extract correlations with both the process parameters and the final performance of the polishing run. Here, in this work, we describe the multiple fronts we have explored in this regard and the results we have gotten so far. This approach promises to have the potential to ultimately provide real-time feedback that can be used to ensure the progress of the run as well as a means for faster optimization.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
NASA NTRS1972

VISCEL: A general-purpose computer program for analysis of linear viscoelastic structures (user's manual), volume 1

This program, an extension of the linear equilibrium problem solver ELAS, is an updated and extended version of its earlier form (written in FORTRAN 2 for the IBM 7094 computer). A synchronized material property concept utilizing incremental time steps and the finite element matrix displacement approach has been adopted for the current analysis. A special option enables employment of constant time steps in the logarithmic scale, thereby reducing computational efforts resulting from accumulative material memory effects. A wide variety of structures with elastic or viscoelastic material properties can be analyzed by VISCEL. The program is written in FORTRAN 5 language for the Univac 1108 computer operating under the EXEC 8 system. Dynamic storage allocation is automatically effected by the program, and the user may request up to 195K core memory in a 260K Univac 1108/EXEC 8 machine. The physical program VISCEL, consisting of about 7200 instructions, has four distinct links (segments), and the compiled program occupies a maximum of about 11700 words decimal of core storage.

Gupta, K. K.↗
NASA NTRS2019

Designing the User Experience for Earth Observation Data Services in the Cloud

NASA's Earth Observation (EO) inventory is projected to grow by an order of magnitude over the next 5-6 years. The current mode for working with EO data of downloading the data to a local machine (laptop, desktop or server) will be difficult to sustain for these upcoming volumes. Therefore, NASA is in the process of developing a capability to host large volume data in commercial cloud, with an eye toward encouraging data analysis in the cloud. However, in order for the user community to take advantage of this new mode of data interaction, the user experience must be redesigned. Cloud-hosted data brings new challenges, such as managing the costs of data egress and working with data in Web Object Storage instead of a Posix filesystem. However, it also brings new opportunities. Scaling data transformation processes may permit more synchronous data services with near-immediate response vs. cumbersome ordering systems with latencies of hours or days. Data co-location in the cloud can facilitate data integration and fusion. Highly scalable filesystems and databases in the cloud support data reorganization to facilitate analysis at scale. In the course of NASA's reimagining of the User Experience for EO data usage relies on end user input gathered through surveys, workshops and meetings (such as this). At the same time, we have embarked on a course of pedagogy and capacity building to help the user community evolve to cloud-based analysis.

Lynnes, Christopher↗
NASA NTRSDate not supplied

Harmonized Sentinel-1 SAR Global River Geometry and Inundation Database

Satellite-based observations on river geometries are sporadic in time, space, or both. Most satellite-based surface water maps, river widths, water surface elevations (WSE), slopes, and bathymetry are asynchronized in time and space. The current configuration of satellites such as Sentinel-6 measured the WSE but is missing the river width, slopes, and depths. To advance hydrological sciences research, there is a need to produce a harmonized time series of river geometry data of non-SWOT satellites in partnership with the upcoming SWOT mission. The SWOT satellite will measure river width, height, and slope but missing river depth measurements in space and time. Further, none of these current satellites measure the WSE, river width, and slopes synchronously. In this work, we use the Sentinel-1 SAR satellite data archive from 2015 to the present to create a global river width and surface water database at the reach scale. A modified version of the Sentinel SAR surface water classification algorithm from ASF is used to quantify the surface water extent on the stream approximately every six days (at the equator) at 10m spatial resolution globally. This 10m water mask is fed into a workflow to quantify the river widths, surface water inundations, slopes, and synthetic bathymetry in SWORD (SWOT River Database) stream networks. A Satellite HAND is used to address the cloud obscured surface water observations using a trained machine learning algorithm. We use WSE derived from the Global Water Monitor from NASA GSFC, Hydroweb from LEGOS, and ICESat-2 to harmonize the WSE observation. And Landsat-8/9 and Sentinel-2 water observations to fill the gaps in the Sentinel-1 SAR database. We use Congo River Basin as a test case where we have more than 500 radar altimetry-based WSE, continuous series of Sentinel-1, ICESat-2, Landsat-8/9, and Sentinel-2 observations. A Congo River hydrologic model is used to generate the streamflow discharge. The satellite observed river reaches are assimilated with the stream flows computed by the routing models. And the downstream reaches in the river network without satellite observations get optimized for discharge/river geometry at each observation cycle. Our final product is a harmonized river geometry dataset (reach's water extent, WSE, slope, synthetic bathymetry) for Congo Basin's SWORD reaches.

Chandana Gangodagamage↗
NASA NTRSDate not supplied

Monocular Ranging for Small Unmanned Aerial Systems in the Far-Field

Recent proliferation of small Unmanned Aerial Systems (sUAS) applications requires onboard collision avoidance systems to mitigate the risk of collision with non-cooperative aircraft and manned aircraft, which may not see sUAS in time to perform an avoidance maneuver. An attractive avenue for onboard collision avoidance is the utilization of machine vision cameras due to their low size, weight and power (SWaP) requirements. In this paper, we characterize the range performance of a machine vision system developed in-house and mounted onto an sUAS. The technique was designed to estimate the performance of a sense-and-avoid system to ensure that the sensing components meet the well-clear requirements for the chosen platform and avoidance strategy. Experimental flight-test data was acquired from test-flights flown along multiple collision geometries for two intruders: a general Aviation (GA) aircraft and a fixed-wing sUAS. The ownship and both intruders were instrumented with inertial navigation systems (INS) recording position and attitude information. The range at first detection, 𝑹𝟎, was extracted from in-flight imagery of head-on collision course geometry synchronized with INS data from both aircraft and ground-truth values extracted from the raw imagery. This initial detection distance, 𝑹𝟎, scales with atmospheric attenuation. Therefore, under clear sky conditions, the derived 𝑹𝟎 value represents the upper bound on the detection range achievable by the test configuration of the detector. Results indicate that the maximum initial detection distance for a 4k resolution action camera fitted with a 41º Field of View (FOV) lens is 2.763 ± 0.037 km for a GA aircraft and 0.881 ± 0.061 km for a fixed-wing sUAS, respectively. The in results this study suggest that a vision-based detect and track system may be analyzed using the sensor characterization and contextualized within aircraft well-clear volumes.

Chester V. Dolph↗
NASA NTRS1983

Actively suspended counter-rotating machine

A counter-rotating machine, such as a positive displacement pump having a pair of meshed, non-contacting helical screws (10,12), subjects its rotating members to axial and radial thrust forces when used for such purposes as compression of liquid or gaseous phase fluids while transporting them through a pump cavity (11,13). Each helical screw (10,12) has a shaft (17,17') which is actively suspended at opposite ends (11a,11b) of the pump cavity by a servo-controlled magnetic bearing assembly (19) and a servo-controlled rotary drive motor (20). Both bearing assemblies and drive motors are mounted on the outside of the pump cavity (11,13). Opto-electric angular position sensors (250) provide synchronization between radial orientation of the drive motors. The bearing assemblies and drive motors conjugately provide axial stabilization and radial centering of the helical screws during volumetric compression of aspirated liquid or gaseous phase fluids.

Studer, Philip A.↗
NASA NTRS1994

The NIST Internet time service

We will describe the NIST Network Time Service which provides time and frequency information over the Internet. Our first time server is located in Boulder, Colorado, a second backup server is under construction there, and we plan to install a third server on the East Coast later this year. The servers are synchronized to UTC(NIST) with an uncertainty of about 0.8 ms RMS and they will respond to time requests from any client on the Internet in several different formats including the DAYTIME, TIME and NTP protocols. The DAYTIME and TIME protocols are the easiest to use and are suitable for providing time to PC's and other small computers. In addition to UTC(NIST), the DAYTIME message provides advance notice of leap seconds and of the transitions to and from Daylight Saving Time. The Daylight Saving Time notice is based on the US transition dates of the first Sunday in April and the last one in October. The NTP is a more complex protocol that is suitable for larger machines; it is normally run as a 'daemon' process in the background and can keep the time of the client to within a few milliseconds of UTC(NIST). We will describe the operating principles of various kinds of client software ranging from a simple program that queries the server once and sets the local clock to more complex 'daemon' processes (such as NTP) that continuously correct the time of the local clock based on periodic calibrations.

Levine, Judah↗
DOE OSTI2022

Lifting and Dropping VMs to Dynamically Transition Between Time- and Space-sharing for Large-Scale HPC Systems

As HPC environments increasingly integrate with edge based systems, system architectures will need to handle a broader class of workloads and scheduling requirements. One result of this shift will be the need to simultaneously support bulk-synchronous parallel (BSP) and on-demand service based applications on the same infrastructure. This in turn will require that future resource management approaches utilize both space-shared as well as time-shared resource scheduling strategies. In this work we introduce the concept of "VM-lifting'' (and its inverse "VM-Dropping'') which allows dynamically switching an HPC workload between space-shared and time-shared scheduling regimes. Our work targets co-kernel based HPC system software environments, in which multiple specialized OS kernels execute natively on dedicated physical resource partitions inside a single compute node. With VM-lifting, a native co-kernel can be migrated at runtime to and from locally hosted Virtual Machine Environments due to changing scheduling requirements of the node. This allows an HPC node to be dynamically (re-)configured as either a time-shared Infrastructure-as-a-Service (IaaS) resource or a dedicated space shared resource based on the current workload demands. We have implemented this approach in the context of the Hobbes Exascale System Software stack and have demonstrated that a node can be reconfigured with minimal impact on the running applications.

Gordon, Nick↗
DOE OSTI2026

Feature-Based PMU Event Classification under Variable PMU Participation and Overlapping Events

Danovo Energy Solution's presented its paper named: Feature-Based PMU Event Classification under Variable PMU Participation and Overlapping Events at the 2026 Georgia Tech Fault & Disturbance Analysis Conference. The full paper can be found at OSTI ID# 3169150 Paper Abstract—Phasor Measurement Units (PMUs) stream time synchronized, high-resolution measurements from the grid, enabling data-driven techniques for event detection and classification. Accurate event classification improves grid reliability and stability. Events can be detected by varying numbers of PMUs and exhibit different durations depending on the event type. This variability challenges standard classifiers that require uniform input sizes. Moreover, multiple events may coincide, which increases classification complexity. Standard classifiers assign each instance to the class with the highest predicted probability, whereas overlapping events may exhibit comparable probabilities across multiple classes. In this study, to handle data size variability, we extract a wide range of time–frequency domain features from all available PMUs for each event into a fixed-length vector, facilitating the application of standard machine learning classifiers, including Random Forest, XGBoost, LightGBM, Support Vector Machine, and Multilayer Perceptron. To account for overlapping events, a probabilistic post-processing step is applied. For a given data instance, if multiple predicted class probabilities exceed 30% and the differences between them are less than 10%, the event is assigned to multiple classes. Experiments using real-world PMU data demonstrate that the Random Forest and XGBoost models achieve the highest accuracy, while the proposed post-processing method yields perfect classification performance on external unseen test sets.

Nematirad, Reza [Danova Energy Solutions]↗
DOE OSTI2025

A Guide to Current Limiting and Stability with Grid-Forming Inverters

The electric power grid is changing. For many decades, the synchronous generator - an electromechanical device invented in 1887 - has been the workhorse and backbone of power grids across the globe. Our entire infrastructure is built around it. But times change, and so does the power grid. Since the early 21st century, we have seen a gradual shift in modern power grids away from synchronous generators to ones dominated by power electronic inverter-based resources (IBRs). Sources such as photovoltaics, wind turbines, battery storage, fuel cells, and other technologies like high-voltage DC transmission interconnections all rely on an inverter to connect and interface with the grid. They are increasingly being installed on the grid to augment, or even replace, traditional energy sources. This change is a fundamental shift that brings tremendous technical challenges and questions: Can a power grid remain stable with many more inverter-based resources? How do we avoid more blackouts on the grid? How do we keep the grid secure and resilient during disturbances? After all, power electronic inverters are nothing like the big, rotating, iron-and-copper machines that the grid heavily relies on. Many of these questions can be answered by using grid-forming (GFM) inverters, yet many research challenges remain. This document explores GFM inverters and how they can help stabilize the future grid, especially during disturbances and contingencies. It summarizes a two-year research and development fellowship program at NREL. We point interested readers to more detailed works developed during the project along the way. Let's dive in

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2026

Feature-Based PMU Event Classification under Variable PMU Participation and Overlapping Events

This paper is the basis for a presentation help at the 2026 Georgia Tech Fault & Disturbance Analysis Conference, which can be found at OSTI # 3168287 Paper Abstract—Phasor Measurement Units (PMUs) stream time synchronized, high-resolution measurements from the grid, enabling data-driven techniques for event detection and classification. Accurate event classification improves grid reliability and stability. Events can be detected by varying numbers of PMUs and exhibit different durations depending on the event type. This variability challenges standard classifiers that require uniform input sizes. Moreover, multiple events may coincide, which increases classification complexity. Standard classifiers assign each instance to the class with the highest predicted probability, whereas overlapping events may exhibit comparable probabilities across multiple classes. In this study, to handle data size variability, we extract a wide range of time–frequency domain features from all available PMUs for each event into a fixed-length vector, facilitating the application of standard machine learning classifiers, including Random Forest, XGBoost, LightGBM, Support Vector Machine, and Multilayer Perceptron. To account for overlapping events, a probabilistic post-processing step is applied. For a given data instance, if multiple predicted class probabilities exceed 30% and the differences between them are less than 10%, the event is assigned to multiple classes. Experiments using real-world PMU data demonstrate that the Random Forest and XGBoost models achieve the highest accuracy, while the proposed post-processing method yields perfect classification performance on external unseen test sets.

Nematirad, Reza [Danovo Energy Solutions]↗
NASA NTRS2017

Sample Processor for Life on Icy Worlds (SPLIce): Design and Test Results

We report the design, development, and testing of the Sample Processor for Life on Icy Worlds (SPLIce) system, a microfluidic sample processor to enable autonomous detection of signatures of life and measurements of habitability parameters in Ocean Worlds. This monolithic fluid processing-and-handling system (Figure 1; mass 0.5 kg) retrieves a 50-L-volume sample and prepares it to supply a suite of detection instruments, each with unique preparation needs. SPLIce has potential applications in orbiter missions that sample ocean plumes, such as found in Saturns icy moon Enceladus, or landed missions on the surface of icy satellites, such as Jupiters moon Europa. Answering the question Are we alone in the universe? is captivating and exceptionally challenging. Even general criteria that define life very broadly include a significant role for water [1,2]. Searches for extinct or extant life therefore prioritize locations of abundant water whether in ancient (Mars), or present (Europa and Enceladus) times. Only two previous planetary missions had onboard fluid processing: the Viking Biology Experiments [3] and Phoenixs Wet Chemistry Laboratory (WCL) [4]. SPLIce differs crucially from those systems, including its capability to process and distribute L-volume samples and the integration autonomous control of a wide range of fluidic functions, including: 1) retrieval of fluid samples from an evacuated sample chamber; 2) onboard multi-year storage of dehydrated reagents; 3) integrated pressure, pH, and conductivity measurement; 4) filtration and retention of insoluble particles for microscopy; 5) dilution or vacuum-driven concentration of samples to accommodate instrument working ranges; 6) removal of gas bubbles from sample aliquots; 7) unidirectional flow (check valves); 8) active flow-path selection (solenoid-actuated valves); 9) metered pumping in 100 nL volume increments. The SPLIce manifold, made of three thermally fused layers of precision-machined cyclo-olefin polymer, supports all fluidic components (Figure 1) and integrated microchannels (125 x 250 m). Fluid is pumped by a stepper-motor-driven pump (Lee Co.). The functionality of the integrated MEMS pressure sensor (Honeywell) and passive check valves (Figure 2) were tested in conjunction with our newly designed integral bubble traps (Figure 3) and hydrophobic membrane-based concentrator (Figure 4). The concentrator (initially tested as a standalone component) demonstrated 5-fold vacuum-evaporative concentration. Polyethylene fused bead beds (PEFBBs; 50 porosity) store drylyophilized buffers, calibrants, and fluorescent dyes, and also promote mixing of sample with calibrant, dye, or H2O. Software-controlled automated tests demonstrated successful 1) fluid delivery to each component 2) valve and pump synchronization 3) sample aliquot delivery to instrument interface ports, and 4) rehydration of vacuum-dried fluorescent dye. In Figure 5, fluorescein on PEFBBs was rehydrated for 15 min using a pump-delivered water aliquot; it is displaced as H2O enters the bottom of the channel and pushes the dye into a check valve. Ultimately, SPLIce will fluorescently label amino acids in the sample for microchip-based electrophoretic (MCE) chiral separation and detection to seek and quantify key organic bio-signatures [5]; it will also deliver sample to a microfluidic version of WCL (mWCL) to measure soluble ions and redox-active species.

Life detection↗
DOE OSTI2021

Global minimization via classical tunneling assisted by collective force field formation

Simple elements interacting in networks can give rise to intricate emergent behaviors. Examples such as synchronization and phase transitions often apply in many contexts, as many different systems may reduce to the same effective model. Here, we demonstrate such a behavior in a model inspired by memristors. When weakly driven, the system is described by movement in an effective potential, but when strongly driven, instabilities cause escapes from local minima, which can be interpreted as an unstable tunneling mechanism. We dub this collective and nonperturbative effect a “Lyapunov force,” which steers the system toward the global minimum of the potential function, even if the full system has a constellation of equilibrium points growing exponentially with the system size. This mechanism is appealing for its physical relevance in nanoscale physics and for its possible applications in optimization, Monte Carlo schemes, and machine learning.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2009

Porting Gravitational Wave Signal Extraction to Parallel Virtual Machine (PVM)

Laser Interferometer Space Antenna (LISA) is a planned NASA-ESA mission to be launched around 2012. The Gravitational Wave detection is fundamentally the determination of frequency, source parameters, and waveform amplitude derived in a specific order from the interferometric time-series of the rotating LISA spacecrafts. The LISA Science Team has developed a Mock LISA Data Challenge intended to promote the testing of complicated nested search algorithms to detect the 100-1 millihertz frequency signals at amplitudes of 10E-21. However, it has become clear that, sequential search of the parameters is very time consuming and ultra-sensitive; hence, a new strategy has been developed. Parallelization of existing sequential search algorithms of Gravitational Wave signal identification consists of decomposing sequential search loops, beginning with outermost loops and working inward. In this process, the main challenge is to detect interdependencies among loops and partitioning the loops so as to preserve concurrency. Existing parallel programs are based upon either shared memory or distributed memory paradigms. In PVM, master and node programs are used to execute parallelization and process spawning. The PVM can handle process management and process addressing schemes using a virtual machine configuration. The task scheduling and the messaging and signaling can be implemented efficiently for the LISA Gravitational Wave search process using a master and 6 nodes. This approach is accomplished using a server that is available at NASA Ames Research Center, and has been dedicated to the LISA Data Challenge Competition. Historically, gravitational wave and source identification parameters have taken around 7 days in this dedicated single thread Linux based server. Using PVM approach, the parameter extraction problem can be reduced to within a day. The low frequency computation and a proxy signal-to-noise ratio are calculated in separate nodes that are controlled by the master using message and vector of data passing. The message passing among nodes follows a pattern of synchronous and asynchronous send-and-receive protocols. The communication model and the message buffers are allocated dynamically to address rapid search of gravitational wave source information in the Mock LISA data sets.

Thirumalainambi, Rajkumar↗
DOE OSTI2020

Deep Generative Models that Solve PDEs: Distributed Computing for Training Large Data-Free Models

Recent progress in scientific machine learning (SciML) has opened up the possibility of training novel neural network architectures that solve complex partial differential equations (PDEs). Several (nearly data free) approaches have been recently reported that successfully solve PDEs, with examples including deep feed forward networks, generative networks, and deep encoder-decoder networks. However, practical adoption of these approaches is limited by the difficulty in training these models, especially to make predictions at large output resolutions (≥1024×1024). Here we report on a software framework for data parallel distributed deep learning that resolves the twin challenges of training these large SciML models - training in reasonable time as well as distributing the storage requirements. Our framework provides several out of the box functionality including (a) loss integrity independent of number of processes, (b) synchronized batch normalization, and (c) distributed higher-order optimization methods. We show excellent scalability of this framework on both cloud as well as HPC clusters, and report on the interplay between bandwidth, network topology and bare metal vs cloud. We deploy this approach to train generative models of sizes hitherto not possible, showing that neural PDE solvers can be viably trained for practical applications. We also demonstrate that distributed higher-order optimization methods are 2-3× faster than stochastic gradient-based methods and provide minimal convergence drift with higher batch-size.

PDEs↗