Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “convergence rate”

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 883 records · Page 49
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Covariance Shaping Over Riemannian Manifolds for Massive MIMO Communication

Acquiring accurate instantaneous channel state information (CSI) is a challenging aspect of massive multi-input multi-output (MIMO) communication. Utilizing statistical information, such as channel covariance matrix, to design statistical beamforming vectors is robust when compared to instantaneous CSI. In this paper, we propose a novel MIMO covariance shaping scheme over Riemannian manifolds. It serves as an effective statistical beamforming solution to a number of close proximity user equipment (UE) that are undergoing substantial channel correlation. Proposed algorithm exploits the Hermitian positive definite nature of covariance matrices lying over Riemannian manifold. We introduce Wasserstein distance function as a Riemannian metric to measure distances between channel covariance matrices. Furthermore, K-means clustering technique is utilized to effectively identify the optimal shape of effective optimal covariance matrices. Our findings suggest that maximizing the geodesic distance between covariance matrices ultimately leads to a corresponding increase in the network throughput, as determined by the beamforming vector used to shape the covariance matrices. Simulation results validate that the proposed solution converges faster than Euclidean-based state-of-the-art, while maintaining the same computational complexity. Finally, the sum rate performance asymptotically achieves full capacity for two-UE case and more than 96% of the upper bound exhaustive search benchmark for multi-UE scenario.

42 ENGINEERING↗
NASA NTRS1993

Rotation and activity among solar-type stars of the Ursa Major Group

We examine rotation and chromospheric activity among G and K dwarfs recently shown to be members of the Ursa Major Group (UMaG). Rotation periods for UMaG stars are smaller than for stars of the same colors in the Hyades, and by an amount corresponding to the Skumanich relation. Most UMaG stars have about the same level of Ca II and K emission, implying that they also have nearly uniform intrinsic rotation rates. That means that the diversity of rotation rates and levels of activity seen among solar-type stars in the Alpha Persei and Pleiades clusters has largely converged by the age of UMaG (0.3 Gyr).

Soderblom, David R.↗
NASA NTRS1997

Global Plate Velocities from the Global Positioning System

We have analyzed 204 days of Global Positioning System (GPS) data from the global GPS network spanning January 1991 through March 1996. On the basis of these GPS coordinate solutions, we have estimated velocities for 38 sites, mostly located on the interiors of the Africa, Antarctica, Australia, Eurasia, Nazca, North America, Pacific, and South America plates. The uncertainties of the horizontal velocity components range from 1.2 to 5.0 mm/yr. With the exception of sites on the Pacific and Nazca plates, the GPS velocities agree with absolute plate model predictions within 95% confidence. For most of the sites in North America, Antarctica, and Eurasia, the agreement is better than 2 mm/yr. We find no persuasive evidence for significant vertical motions (less than 3 standard deviations), except at four sites. Three of these four were sites constrained to geodetic reference frame velocities. The GPS velocities were then used to estimate angular velocities for eight tectonic plates. Absolute angular velocities derived from the GPS data agree with the no net rotation (NNR) NUVEL-1A model within 95% confidence except for the Pacific plate. Our pole of rotation for the Pacific plate lies 11.5 deg west of the NNR NUVEL-1A pole, with an angular speed 10% faster. Our relative angular velocities agree with NUVEL-1A except for some involving the Pacific plate. While our Pacific-North America angular velocity differs significantly from NUVEL-1A, our model and NUVEL-1A predict very small differences in relative motion along the Pacific-North America plate boundary itself. Our Pacific-Australia and Pacific- Eurasia angular velocities are significantly faster than NUVEL-1A, predicting more rapid convergence at these two plate boundaries. Along the East Pacific Pise, our Pacific-Nazca angular velocity agrees in both rate and azimuth with NUVFL-1A.

Larson, Kristine M.↗
NASA NTRS1999

A Comparison of Three Navier-Stokes Solvers for Exhaust Nozzle Flowfields

A comparison of the NPARC, PAB, and WIND (previously known as NASTD) Navier-Stokes solvers is made for two flow cases with turbulent mixing as the dominant flow characteristic, a two-dimensional ejector nozzle and a Mach 1.5 elliptic jet. The objective of the work is to determine if comparable predictions of nozzle flows can be obtained from different Navier-Stokes codes employed in a multiple site research program. A single computational grid was constructed for each of the two flows and used for all of the Navier-Stokes solvers. In addition, similar k-e based turbulence models were employed in each code, and boundary conditions were specified as similarly as possible across the codes. Comparisons of mass flow rates, velocity profiles, and turbulence model quantities are made between the computations and experimental data. The computational cost of obtaining converged solutions with each of the codes is also documented. Results indicate that all of the codes provided similar predictions for the two nozzle flows. Agreement of the Navier-Stokes calculations with experimental data was good for the ejector nozzle. However, for the Mach 1.5 elliptic jet, the calculations were unable to accurately capture the development of the three dimensional elliptic mixing layer.

Georgiadis, Nicholas J.↗
NASA NTRS2004

Review of TRMM/GPM Rainfall Algorithm Validation

A review is presented concerning current progress on evaluation and validation of standard Tropical Rainfall Measuring Mission (TRMM) precipitation retrieval algorithms and the prospects for implementing an improved validation research program for the next generation Global Precipitation Measurement (GPM) Mission. All standard TRMM algorithms are physical in design, and are thus based on fundamental principles of microwave radiative transfer and its interaction with semi-detailed cloud microphysical constituents. They are evaluated for consistency and degree of equivalence with one another, as well as intercompared to radar-retrieved rainfall at TRMM's four main ground validation sites. Similarities and differences are interpreted in the context of the radiative and microphysical assumptions underpinning the algorithms. Results indicate that the current accuracies of the TRMM Version 6 algorithms are approximately 15% at zonal-averaged / monthly scales with precisions of approximately 25% for full resolution / instantaneous rain rate estimates (i.e., level 2 retrievals). Strengths and weaknesses of the TRMM validation approach are summarized. Because the dew of convergence of level 2 TRMM algorithms is being used as a guide for setting validation requirements for the GPM mission, it is important that the GPM algorithm validation program be improved to ensure concomitant improvement in the standard GPM retrieval algorithms. An overview of the GPM Mission's validation plan is provided including a description of a new type of physical validation model using an analytic 3-dimensional radiative transfer model.

Smith, Eric A.↗
NASA NTRS2017

Viscous Three-Dimensional Simulation of Flow in an Axial Low Pressure Compressor at Engine Icing Operating Points

Viscous three-dimensional simulations of the Honeywell ALF502R-5 low pressure compressor (sometimes called a booster) using the NASA Glenn code GlennHT have been carried out. A total of ten simulations were produced. Five operating points are investigated, with each point run with two different wall thermal conditions. These operating points are at, or near, points where engine icing has been determined to be likely. In the future, the results of this study will be used for further analysis such as predicting collection efficiency of ice particles and ice growth rates at various locations in the compressor. A mixing plane boundary condition is used between each blade row, resulting in convergence to steady state within each blade row. The k-omega turbulence model of Wilcox, combined with viscous grid spacing near the wall on the order of one, is used to resolve the turbulent boundary layers. For each of the operating points, heat transfer coefficients are generated on the blades and walls. The heat transfer coefficients are produced by running the operating point with two different wall thermal conditions and then solving simultaneously for the heat transfer coefficient and adiabatic wall temperature at each point. Average Nusselt numbers are calculated for the most relevant surfaces. The values are seen to scale with Reynolds number to approximately a power of 0.7. Additionally, images of surface distribution of Nusselt number are presented. Qualitative comparison between the five operating points show that there is relatively little change in the character of the distribution. The dominant observed effect is that of an overall scaling, which is expected due to Reynolds number differences. One interesting aspect about the Nusselt number distribution is observed on the casing (outer diameter) downstream of the exit guide vanes (EGVs). The Nusselt number is relatively high between the pairs of EGVs, with two lower troughs downstream of each EGV trailing edge. This is of particular interest since rather complex ice shapes have been observed in that region.

compressor↗
NASA NTRSDate not supplied

Some Aeronautical Communications Experiments

Classically there has existed an asymmetry between the computing and communicating sides of aerospace systems. Over the past few decades, this asymmetry has shifted to favoring communication link technologies, meaning that advancements in available central processing units (CPUs), storage devices, and internal data buses have stagnated. Indeed, the increased emphasis placed on refining subsystem performance such as with antenna bandwidth in phased arrays, amplifier power efficiency, software defined radio (SDR) flexibility and encoding for data compression and error correction has given rise to successful debuts of multi-gigabit-per-second data return from long space-link distances. These accomplishments are easily quantifiable through link budgets and illustrate what is possible, but also reveal the deficiencies in overall communications capabilities. In particular, the ever-accelerating presence of aerospace vehicles gives rise to newer and larger classes of challenges to address the needs of 21st century systems. Furthermore remote sensing and imaging capabilities have far outpaced our ability to transmit their products to the ground, so we are increasingly dependent on pre-processing and downselection to contend with the communications bottleneck. No longer may we depend upon the constrained logistics in delivering end-to-end data delivery through manual reconfigurations, static event scheduling and execution on a per-vehicle basis, for these methods do not scale and therefore must give way to dynamic, networked approaches with an overall systems view in mind. Emerging mission requirements exhibit a trend toward multiple smaller-scale vehicles working together to perform dissimilar observations. Such operations necessitate sensor fusion across a constellation, and where data processing may be distributed throughout a fairly disconnected network whose topology changes over time in non-deterministic manners. Individual communications link performance is still very relevant to deploying an effective communications system, but now must be embedded within a greater architecture of capability to optimally utilize the bandwidth available from each link to generate an ultimate end-to-end quality of service. The deleterious effects of timing uncertainty across the arrangement presents a challenge to measurement synchronization and delivery, so a successful deployed system needs to be tolerant to the delays inherent in time-of-light between elements and digital processing latencies existing at each node. In this presentation we share the flight test results from a high performance Gbps laser communications terminal evaluated with a suite of store and forward capabilities called High-rate Delay Tolerant Networking (HDTN). The communications payload is operated over Lake Erie across a range of configurations including several convergence layers, and is evaluated to determine recovery time after link disruptions, information loss, efficiency and speed. The effectiveness of utilizing a flying laboratory to increase the Technology Readiness Level (TRL) of an integrated system in relevant environments is discussed, as well as the value of conducting aeronautics experiments to retire risk for technology infusion into space missions. Upcoming flight campaigns will be presented, including opportunities to demonstrate secure command and control, data intensive hyperspectral imaging, quantum link characterization, 4k High Definition (HD) video streaming and internetworked space-ground-aero relay operations. These experiments will pave the way for future missions which will depend upon interoperability across disparate government and privately owned networks, involve contention with uncertain and dynamic timing, and require agility to autonomously configure optimal parameters across networks of ever-increasing size and complexity to ensure data delivery. https://www1.grc.nasa.gov/space/scan/acs/tech-studies/dtn/

Daniel Raible↗
NASA NTRS1992

Sea level variation

Published values for the long-term, global mean sea level rise determined from tide gauge records range from about one to three mm per year. The scatter of the estimates appears to arise largely from the use of data from gauges located at convergent tectonic plate boundaries where changes of land elevation give fictitious sea level trends, and the effects of large interdecadal and longer sea level variations on short (less than 50+ years) or sappy records. In addition, virtually all gauges undergo subsidence or uplift due to isostatic rebound from the last deglaciation at a rate comparable to or greater than the secular rise of sea level. Modeling rebound by the ICE-3G model of Tushingham and Peltier (1990) and avoiding tide gauge records in areas of converging tectonic plates produces a highly consistent set of long sea level records. A global set of 21 such stations in nine oceanic regions with an average record length of 76 years during the period 1880-1980 yields the global sea level rise value 1.8 mm/year +/- 0.1. Greenhouse warming scenarios commonly forecast an additional acceleration of global sea level in the next 5 or 6+ decades in the range 0.1-0.2 mm/yr2. Because of the large power at low frequencies in the sea level spectrum, very long tide gauge records (75 years minimum) have been examined for past apparent sea level acceleration. For the 80-year period 1905-1985, 23 essentially complete tide gauge records in 10 geographic groups are available for analysis. These yielded the apparent global acceleration -0.011 (+/- 0.012) mm/yr2. A larger, less uniform set of 37 records in the same 10 groups with 92 years average length covering the 141 years from 1850-1991 gave 0.001 (+/- 0.008) mm/yr2. Thus there is no evidence for an apparent acceleration in the past 100+ years that is significant either statistically, or in comparison to values associated with global warming. Unfortunately, the large interdecadal fluctuations of sea level severely affect estimates of global sea level acceleration for time spans of less than about 50 years. This means that tide gauges alone cannot serve as a reliable leading indicator of climate change in less than many decades. This time required can be significantly reduced if the interdecadal fluctuations of sea level can be understood in terms of their forcing mechanisms, and then removed from the tide gauge records.

Douglas, Bruce C.↗
DOE OSTI2025

Local practically safe extremum seeking with assignable rate of attractivity to the safe set

We present Assignably Safe Extremum Seeking (ASfES), an algorithm designed to minimize a measured, static objective function while maintaining a measured, static metric of safety (a control barrier function or CBF) to be positive in a practical sense. We ensure that for trajectories with safe initial conditions, the violation of safety can be made arbitrarily small through appropriately chosen design constants. We also guarantee an assignable “attractivity” rate: from unsafe initial conditions, the trajectories approach the safe set, in the sense of the measured CBF, at a rate no slower than a user-assigned rate. Similarly, from safe initial conditions, the trajectories approach the unsafe set, in the sense of the CBF, no faster than the assigned attractivity rate. The feature of assignable attractivity is not present in the semiglobal version of safe extremum seeking, where the semiglobality of convergence is achieved by slowing the adaptation. We also demonstrate local convergence of the parameter to a neighborhood of the minimum of a quadratic objective function constrained to the safe set with a linear CBF. The ASfES algorithm and analysis are multivariable, but we also extend the algorithm to a Newton-Based ASfES scheme which we show is only useful in the scalar case. The proven properties of the designs are illustrated through simulation examples.

42 ENGINEERING↗
DOE OSTI2022

Weak-Bonding Elements Lead to High Thermoelectric Performance in BaSnS 3 and SrSnS 3 : A First-Principles Study

SnS2, an earth-abundant and ecofriendly material, is limited as a thermoelectric material because of the high lattice thermal conductivity κ L and low carrier mobility μ. By introducing weak-bonding elements Ba or Sr into the SnS 2 framework, we discovered two SnS 2 -based materials BaSnS3 and SrSnS3 with the calculated low κL values of 0.15 and 0.17 W m -1 K -1 , respectively, along the a-axis. The low group velocity and high lattice anharmonicity originating from the weakened and distorted Sn–S bonding network are found in both systems. Moreover, the vibrations of Ba and Sr induce low-lying optical phonons, which strongly couple with the acoustic phonons and strengthen the phonon scattering rates. Compared to SnS 2 , both compounds present lower single-band effective masses, smaller deformation potential constants, and better band convergence, which enhance μ with an insignificantly reduced effective mass. By solving the linearized Boltzmann transport equation with a nonempirical carrier lifetime, we predict excellent ZT values of 2.89 and 2.77 along the a-axis at 900 K in BaSnS 3 and SrSnS 3 , respectively. Further phase diagram calculations of Ba 1–x Sr x SnS 3 solid solutions propose a new compound, Ba 0.5 Sr 0.5 SnS 3 , with an even higher ZT of 3.0. Our work analyzes explicitly how weak-bonding elements enhance μ and suppress κL simultaneously in SnS 2 -analogous systems with a series of compounds nominated as potential high-performance thermoelectric materials.

36 MATERIALS SCIENCE↗
DOE OSTI2020

Calculation of electron-ion temperature equilibration rates and friction coefficients in plasmas and liquid metals using quantum molecular dynamics

In this work, we discuss a method to calculate with quantum molecular dynamics simulations the rate of energy exchanges between electrons and ions in two-temperature plasmas, liquid metals, and hot solids. Promising results from this method were recently reported for various materials and physical conditions. Like other ab initio calculations, the approach offers a very useful comparison with the experimental measurements and permits an extension into conditions not covered by the experiments. The energy relaxation rate is related to the friction coefficients felt by individual ions due to their nonadiabatic interactions with electrons. Each coefficient satisfies a Kubo relation given by the time integral of the autocorrelation function of the interaction force between an ion and the electrons. These Kubo relations are evaluated using the output of quantum molecular dynamics calculations in which electrons are treated in the framework of finite-temperature density functional theory. The calculation presents difficulties that are unlike those encountered with the Kubo formulas for the electrical and thermal conductivities. In particular, the widely used Kubo-Greenwood approximation is inapplicable here. Indeed, the friction coefficients and the energy relaxation rate diverge in this approximation since it does not properly account for the electronic screening of electron-ion interactions. The inclusion of screening effects considerably complicates the calculations. We discuss the physically motivated approximations we applied to deal with these complications in order to investigate a widest range of materials and physical conditions. Unlike the standard method used for the electronic conductivities, the Kubo formulas are evaluated directly in the time domain and not in the energy domain, which spares one from needing to introduce an extraneous undetermined numerical parameter to account for the discrete character of the numerical density of states. We highlight interesting properties of the energy relaxation rate not shared by other electronic properties, in particular its self-averaging character. We then present a detailed parametric and convergence study with the numerical parameters, including the system size, the number of bands and k-points, and the physical approximations for the dielectric function and the exchange-correlation energy.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
NASA NTRS1980

Deep convection on day 261 of GATE

The structural features of the deep convection observed on September 18, 1974, day 261 of the GARP Atlantic Tropical Experiment (GATE), as the ridge axis of a 700 mb wave passed the center of the GATE B-scale network are reported. Satellite and aircraft maps indicate the presence of clouds penetrating above 2.5 km into the middle troposphere organized in bands about 9 km apart and aligned roughly along the direction of the wind shear in the cloud layer. Radar echoes corresponding to cumulus convection of lifetime, peak height and peak rainfall rates on the orders of 30 min, 6 km and 1.3 mm/h, respectively, were observed to triple in number density as convergence at 950 hPa increased from 1.5 to 3 x 10 to the -5th/sec. The structural features of the radar echoes indicate that the day was similar to a mesoscale precipitation feature of Leary and Houe (1979), with the cluster consisting of many echoes appearing in succession. Data from aircraft penetrations of the deep convection reveal downdrafts accompanying the precipitation and updrafts immediately to their south. Shipboard and rawinsonde observations show that the convective downdrafts brought down air of low pseudo-equivalent potential temperature, with local surface convergence of up to 0.001/sec. Mean wind shears through the cloud layer to the top of a main cloud layer are found to be only 75% greater than those of Malkus (1958) for the Caribbean, with shears just above the cloud base several factors larger.

Warner, C.↗
NASA NTRS1989

Sources of Sahel precipitation for simulated drought and rainy seasons

The sources of sub-Saharan precipitation are studied using diagnostic procedures integrated into the code of the GISS climate model. Water vapor evaporating from defined source regions is 'tagged', allowing the determination of the relative contributions of each evaporative source to the simulated July rainfall in the Sahel. Two June-July simulations are studied to compare the moisture sources, moisture convergence patterns and the spatial variations of precipitation for rainy and drought conditions. Results for this case study indicate that patterns of moisture convergence and divergence over northern Africa had a stronger influence on model rainfall over the sub-Sahara than did evaporation rates over the adjacent oceans or moisture advection from ocean to continent. While local continental evaporation contributed significant amounts of water to Sahelian precipitation in the'rainy' simulation, moisture from the Indian Ocean did not precipitate over the Sahel in either case.

Druyan, Leonard M.↗
NASA NTRS1992

The structure of intense vorticity in homogeneous isotropic turbulence

The structure of the intense vorticity regions is studied in numerically simulated homogeneous, isotropic, equilibrium turbulent flow fields at four different Reynolds numbers in the range Re(sub lambda) = 36-171. In accordance with previous investigators, this vorticity is found to be organized in coherent, cylindrical or ribbon-like, vortices ('worms'). A statistical study suggests that they are just especially intense features of the background, O(omega'), vorticity. Their radii scale with the Kolmogorov microscale and their lengths with the integral scale of the flow. An interesting observation is that the Reynolds number based on the circulation of the intense vortices, gamma/nu, increases monotonically with Re(sub lambda), raising the question of the stability of the structures in the limit of Re(sub lambda) approaching infinity. One and two-dimensional statistics of vorticity and strain are presented; they are non-gaussian, and the behavior of their tails depends strongly on the Reynolds number. There is no evidence of convergence to a limiting distribution in our range of Re(sub lambda), even though the energy spectra and the energy dissipation rate show good asymptotic properties in the higher Reynolds number cases. Evidence is presented to show that worms are natural features of the flow and that they do not depend on the particular forcing scheme.

Jimenez, J.↗
DOE OSTI2024

A Simulated Evaluation of Powder Flowability Through a Partially Obstructed Consumable in Blown Powder Directed Energy Deposition Systems

Abstract In the interest of continued industrialization of metal additive manufacturing in modern production environments, cost is often referenced as a primary deterrent to new adopters. Conventional economic models for additive systems, processes, and supply chains often focus on specific process applications with little generalizability, or they neglect significant costs associated with production such as machine maintenance and consumable part replacement. Compounding the latter issue are substantial knowledge gaps in consumable part wear characterization for additive and other convergent manufacturing systems. In coaxial blown powder directed energy deposition systems, gas atomized metal powder is wasted during material deposition at a rate that is partly dependent on present wear phenomena in a consumable nozzle housed in the cladding head assembly. The price and lead time required to replace the nozzle incentivizes its reuse even when visibly worn. Often this initiates a process quality decline in the form of underbuilt geometry and internal defects due to losses in powder catchment efficiency. While depositing H13 steel using a hybrid manufacturing machine tool equipped with such a deposition system, a unique partial clog with a bridge-like structure formed at the consumable nozzle exit when supporting argon gas flows failed mid-process. To further understand coaxial multi-phase powder flow in the event of support gas failure, a computational fluid dynamics simulation is tailored to relevant process parameters, H13 powder material profile, and machine operator observations collected after the incident. The resulting differences in powder flow compared to control gas flow parameters is presented and discussed. The powder flowability and performance of the clogged nozzle is then assessed by using an optical profilometer to extract the profile of the clog and recreate the clog geometry within the simulation environment. In past work this simulation has been experimentally validated for a 316L steel powder material profile and used specifically for analyzing powder stream geometry and catchment efficiency. After the initial powder flow characterization, the clog is removed, and the nozzle is reprofiled. After removing the obstructing clog, the newly unobstructed nozzle geometry, the original off the shelf nozzle geometry, and additional nozzle profiles exploring different consumable refurbishment strategies are reevaluated in the simulation. Powder catchment efficiency for all variant nozzle geometries and relevant flow variables are compared and discussed, along with potential mitigation strategies for optimizing powder flowability with worn consumables. This work expands on the known morphology of blown powder obstructions and wear defects present in consumable coaxial nozzles while discussing pragmatic simulation driven responses to unanticipated subsystem failure in hybrid manufacturing machining platforms.

DeWitte, Lisa↗
NASA NTRS1979

Physical models of Comet Halley based upon qualitative data from the 1910 apparition

A semi-empirical theory was developed, tying the hydrogen production rate of a comet to its visual light curve. Two unknown constants, which are essentially the efficiencies of these two processes, are derived from the comets for which quantitative data exist. This theory has been applied to P/Halley, using the light curve of Yeomans, which was derived from the original 1910 observations, with aperture corrections applied. Given the hydrogen production rate, it is assumed that N(H20) = 0.5 N(H) and that there is 0.2 N(H20) of other species. From several converging lines of evidence the mass of dust is taken as half the mass of gas. The Sekanina-Miller distribution function of particle size is used. Particle velocities are calculated using Delsemmes version of the Probstein theory. From these, near nucleus dust density, flux, etc. are calculated. Dust envelope sizes are calculated from the old mechanical theory of tails. Extreme models are calculated from the old mechanical theory of tails. Extreme models are also furnished based upon a "sum-of-negative-tolerance" approach.

Newburn, R. L.↗
NASA NTRS2015

Tropical Gravity Wave Momentum Fluxes and Latent Heating Distributions

Recent satellite determinations of global distributions of absolute gravity wave (GW) momentum fluxes in the lower stratosphere show maxima over the summer subtropical continents and little evidence of GW momentum fluxes associated with the intertropical convergence zone (ITCZ). This seems to be at odds with parameterizations forGWmomentum fluxes, where the source is a function of latent heating rates, which are largest in the region of the ITCZ in terms of monthly averages. The authors have examined global distributions of atmospheric latent heating, cloud-top-pressure altitudes, and lower-stratosphere absolute GW momentum fluxes and have found that monthly averages of the lower-stratosphere GW momentum fluxes more closely resemble the monthly mean cloud-top altitudes rather than the monthly mean rates of latent heating. These regions of highest cloud-top altitudes occur when rates of latent heating are largest on the time scale of cloud growth. This, plus previously published studies, suggests that convective sources for stratospheric GW momentum fluxes, being a function of the rate of latent heating, will require either a climate model to correctly model this rate of latent heating or some ad hoc adjustments to account for shortcomings in a climate model's land-sea differences in convective latent heating.

momentum↗
DOE OSTI2024

Geometry-aware training of factorized layers in tensor Tucker format

Reducing parameter redundancies in neural network architectures is crucial for achieving feasible computational and memory requirements during train and inference of large networks. Given its easy implementation and flexibility, one promising approach is layer factorization, which reshapes weight tensors into a matrix format and parameterizes it as the product of two rank-r matrices. However, this family of approaches often requires an initial full-model warm-up phase, prior knowledge of a feasible rank, and it is sensitive to parameter initialization.In this work, we introduce a novel approach to train the factors of a Tucker decomposition of the weight tensors. Our training proposal proves to be optimal in locally approximating the original unfactorized dynamics and stable for the initialization. Furthermore, the rank of each mode is dynamically updated during training.We provide a theoretical analysis of the algorithm, showing convergence, approximation and local descent guarantees. The method's performance is further illustrated through a variety of experiments, showing remarkable training compression rates and comparable or even better performance than the full baseline and alternative layer factorization strategies.

Zangrando, Emanuele [Gran Sasso Science Institute ↗