Search NASA⌕ Search

SEARCH · Search NASA

Results for “TENSOR ANALYSIS”

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
All sourcesNASA NTRSDOE OSTI
DOE OSTI2020

Normal modes in thermal AdS via the Selberg zeta function

The heat kernel and quasinormal mode methods of computing 1-loop partition functions of spin s s fields on hyperbolic quotient spacetimes \mathbb{H}^{3}/\mathbb{Z} ℍ 3 / ℤ are related via the Selberg zeta function. We extend that analysis to thermal \text{AdS}_{2n+1} AdS 2 n + 1 backgrounds, with quotient structure \mathbb{H}^{2n+1}/\mathbb{Z} ℍ 2 n + 1 / ℤ . Specifically, we demonstrate the zeros of the Selberg function encode the normal mode frequencies of spin fields upon removal of non-square-integrable modes. With this information we construct the 1-loop partition functions for symmetric transverse traceless tensors in terms of the Selberg zeta function and find exact agreement with the heat kernel method.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2020

Physics Division Strategic Plan Fiscal Years 2020-2024

The vision of the Physics Division (PHY) at Argonne National Laboratory is to continue enhancing its role as a world-leading institution in basic nuclear physics research and its applications. Key to this vision is for PHY to continue to safely and effectively operate and evolve the capabilities of the Argonne Tandem Linac Accelerator System (ATLAS) facility to best serve its users. ATLAS is the Department of Energy (DOE) accelerator facility for low-energy nuclear physics research. The research carried out by PHY covers many themes in contemporary science but can be distilled into five areas of focus. These themes are interconnected, with continuously evolving synergies between the various groups and facilities in PHY and the broader Laboratory. This five-year strategic plan serves to illustrate our current capabilities and new directions related to these five areas of focus (ATLAS also develops an independent strategic plan): Accelerator research and design. The goal of this theme is to design, fabricate, test, and implement novel accelerator systems, with a focus on high-intensity ion and electron systems. These R&D activities have led to enhancements in the ATLAS accelerator system. Among others, the division’s accelerator systems are in use or planned for use at Fermilab, the Advanced Photon Source, and in a broad variety of applications at the Facility for Rare Isotope Beams (FRIB). Atom trapping and fundamental symmetries. The goal of this theme is to explore and exploit the uses of advanced laser cooling and trapping techniques to manipulate atoms. There are three main areas of focus. First is in the application of the atom trap trace analysis (ATTA) technique for age determination of groundwater and ice by using radio-krypton dating. The next two trapping-based programs involve tests of fundamental symmetries in nature: the cooling and trapping of radium-225 with the aim of determining limits on an observation of its electric dipole moment and precision measurements of the beta decay properties of helium-6 to set limits on the tensor coupling constant. This is complemented by measurements of similar properties in lithium-8 and boron-8. Nuclear astrophysics. The goal of this theme is to enhance our understanding of how elements are created in the universe via explosive nucleosynthesis and how stars evolve. To meet the challenges of this theme, many of the capabilities of ATLAS are being enhanced, including the development of new beams through a new in-flight separator (RAISOR) and the anticipated neutron-generator upgrade of the Californium Rare Isotope Breeder Upgrade facility (nuCARIBU). These are coupled to state-of-the-art instruments such as the Canadian Penning Trap, Helical Orbit Spectrometer, Gammasphere, GRETINA, Multi-Sampling Ionization Chamber, the Fragment Mass Analyzer, and Argonne Gas-filled Fragment Analyzer, and a new low-background experimental area for decay studies. Nuclear structure. The goal of this theme is to understand the structure of nuclei, both stable and radioactive, in terms of single-particle properties, their shapes, and the dynamics governing reactions between them. These include questions such as what are the limits of nuclear stability and what are the properties of super heavy nuclei. As with the nuclear astrophysics theme, the capabilities of ATLAS, guided by the ATLAS user community, are continuously being enhanced to this end. The Division has strategic initiatives to play a leading role in the development of instrumentation and research programs at the FRIB. Quantum chromodynamics (QCD) and hadron physics. The goal of this theme is to lead major research programs focused on revealing the quark and gluon structure of protons, neutrons, nuclei, and short-lived mesons and baryons. These involve major programs at Jefferson Lab, Fermilab, and smaller facilities. They are complemented by theoretical endeavors centered on the question of how hadrons and their properties emerge from QCD. Out of these activities arise strategic initiatives to play a major role in the forthcoming Electron-Ion Collider (EIC).

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2021

Demonstration of CFD to support the coupled analysis of a reactor pressure vessel subjected to pressurized thermal shock

The structural components that comprise nuclear reactors and their supporting structures are subjected to harsh operating environments that can challenge their integrity, especially after exposure for extended duration or under accident condition. As one of the most significant components of a reactor, the Reactor Pressure Vessel (RPV) is exposed to an aggressive environment during the operation time (e.g. more than 40 years). Aging degradation mechanisms (e.g. thermo-fatigue) could grow initial defects up to a critical size, increasing the susceptibility to failure in the RPV. The conventional methods are mostly based on simple crack and structure geometries. Very limited studies consider the real conditions of the RPV subjected to a thermal shock due to a Loss of Coolant Accident (LOCA). During a LOCA event, the most severe conditions take place when the emergency core cooling (ECC) water is injected inside the cold legs filled initially with hotter water and/or steam. The rapid cooling of the down-comer and the internal RPV surface followed probably by re-pressurization of the RPV causes large temperature gradients and variation of pressure which induces thermal-mechanical stresses. In order to develop the model for integrity assessment of a reactor pressure vessel (RPV) subjected to pressurized thermal shock (PTS), a multi-physics simulation, which includes the thermo-hydraulic, thermo-mechanical and fracture mechanics analyses is necessary. The prediction of the temperature field is achieved by using computational fluid dynamics (CFD) simulation. In this report, a demonstration CFD standalone simulation is performed to support coupled analysis for Reactor Pressure Vessel (RPV) subjected to Pressurized Thermal Shock (PTS). The study use a simplified computational domain to represents a real RPV. The purpose of the study is to demonstrate the transient temperature response of RPV to ECC injection. The CFD model is built in a robust and efficient way for further coupled calculation. The next steps of this work, including the coupled thermal and tensor mechanics capabilities using Cardinal are expected to be complete by the end of FY21 for the demo problem. After this, into FY22, the capability will be demonstrated for a realistic RPV.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2023

Planck constraints and gravitational wave forecasts for primordial black hole dark matter seeded by multifield inflation

Here, we perform a Markov Chain Monte Carlo analysis of a simple yet generic multifield inflation model characterized by two scalar fields coupled to each other and nonminimally coupled to gravity, fit to Planck 2018 cosmic microwave background (CMB) data. In particular, model parameters are constrained by data on the amplitude of the primordial power spectrum of scalar curvature perturbations on CMB scales A s , the spectral index n s , and the ratio of power in tensor to scalar modes r, with a prior that the primordial power spectrum should also lead to primordial black hole (PBH) production sufficient to account for the observed dark matter abundance. We find that n s in particular controls the constraints on our model. Whereas previous studies of PBH formation from an ultraslow-roll phase of inflation have highlighted the need for at least one model parameter to be highly fine-tuned, we identify a degeneracy direction in parameter space such that shifts by ~10% of one parameter can be compensated by comparable shifts in other parameters while preserving a close fit between model predictions and observations. Furthermore, we find this allowed parameter region produces observable gravitational wave signals in the frequency ranges to which upcoming experiments are projected to be sensitive, including Advanced LIGO and Virgo, the Einstein Telescope, Cosmic Explorer, DECIGO, and LISA.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2025

Performance evaluations of signed and unsigned noisy approximate quantum Fourier arithmetic

The Quantum Fourier Transform (QFT) grants competitive advantages, especially in resource usage and circuit approximation, for performing arithmetic operations on quantum computers, and offers a potential route toward a numerical quantum-computational paradigm. In this paper, we utilize efficient techniques to implement QFT-based integer addition and multiplications. These operations are fundamental to various quantum applications including Shor’s algorithm, weighted-sum optimization problems in data processing and machine learning, and quantum algorithms requiring inner products. We carry out performance evaluations of these implementations based on IBM’s superconducting-qubit architecture using different compatible noise models. We isolate the sensitivity of the component quantum circuits on both one-/two-qubit gate error rates, and the number of the arithmetic operands’ superposed integer states. We analyze performance and identify the most effective approximation depths for unsigned quantum addition and quantum multiplication within the given context. We then perform a similar analysis of signed addition and compare to the unsigned results. We observe significant dependency of the optimal approximation depth on the degree of machine noise and the number of superposed states in certain performance regimes. Finally, we elaborate on the algorithmic challenges—relevant to signed, unsigned, modular and non-modular versions—that could also be applied to current implementations of QFT-based subtraction, division, exponentiation, and their potential tensor extensions. Here, we analyze the performance trends in our results and speculate on possible future developments within this computational paradigm.

Computational models↗
DOE OSTI2020

Oscillating scalar fields and the Hubble tension: A resolution with novel signatures

Here, we present a detailed investigation of a subdominant oscillating scalar field [“early dark energy” (EDE)] in the context of resolving the Hubble tension. Consistent with earlier work, but without relying on fluid approximations, we find that a scalar field frozen due to Hubble friction until log 10 ($z_c$) ~ 3.5, reaching $ρ_{\text{EDE}}(z_c)/ρ_{\text{tot}}$ ~ 10 % and diluting faster than matter afterwards, can bring cosmic microwave background (CMB), baryonic acoustic oscillations, supernovae luminosity distances, and the late-time estimate of the Hubble constant from the SH0ES Collaboration into agreement. A scalar field potential that scales as $V (\phi) \propto \phi ^{2n}$ with 2 ≲ $\textit{n}$ ≲ 3.4 around the minimum is preferred at the 68% confidence level, and the Planck polarization places additional constraints on the dynamics of perturbations in the scalar field. In particular, the data prefer a potential that flattens at large field displacements. A Markov-chain Monte Carlo analysis of mock data shows that the next-generation CMB observations (i.e., CMB-S4) can unambiguously detect the presence of the EDE at a very high significance. This projected sensitivity to the EDE dynamics is mainly driven by improved measurements of the E-mode polarization. We also explore new observational signatures of EDE scalar field dynamics: (i) We find that depending on the strength of the tensor-to-scalar ratio, the presence of the EDE might imply the existence of isocurvature perturbations in the CMB. (ii) We show that a strikingly rapid, scale-dependent growth of EDE field perturbations can result from parametric resonance driven by the anharmonic oscillating field for $\textit{n}$ ≈ 2. This instability and ensuing potentially nonlinear, spatially inhomogeneous, dynamics may provide unique signatures of this scenario.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2021

Complex Fault Geometry of the 2020 M ww 6.5 Monte Cristo Range, Nevada, Earthquake Sequence

On 15 May 2020 an M ww 6.5 earthquake occurred beneath the Monte Cristo Range in the Mina Deflection region of western Nevada. Rapid deployment of eight temporary seismic stations enabled detailed analysis of its productive and slowly decaying aftershock sequence (p=0.8), which included ~18,000 autodetected events in 3.5 months. Double-difference, waveform-based relative relocation of 16,714 earthquakes reveals a complex network of faults, many of which cross the inferred 35-km-long east–northeast-striking, left-lateral mainshock rupture. Seismicity aligns with left-lateral, right-lateral, and normal mechanism moment tensors of 128 of the largest earthquakes. The mainshock occurred near the middle of the aftershock zone at the intersection of two distinct zones of seismicity. In the western section, numerous subparallel, shallow, north-northeast-striking faults form a broad flower-structure-like fault mesh that coalesces at depth into a near-vertical, left-lateral fault. We infer the near-vertical fault to be a region of significant slip in the mainshock and an eastward extension of the left-lateral Candelaria fault. Near the mainshock hypocenter, seismicity occurs on a northeast-striking, west-dipping structure that extends north from the eastern Columbus Salt Marsh normal fault. Together, these two intersecting structures bound the Columbus Salt Marsh tectonic basin. East of this intersection and the mainshock hypocenter, seismicity occurs in a narrow, near-vertical, east-northeast-striking fault zone through to its eastern terminus. At the eastern end, the aftershock zone broadens and extends northwest toward the southern extension of the northwest-striking, right-lateral Petrified Springs fault system. The eastern section hosts significantly fewer aftershocks than the western section, but has more moment release. We infer that shallow aftershocks throughout the system highlight fault-fracture meshes that connect mapped fault systems at depth. Here, comparing earthquake data with surface ruptures and a simple geodetic fault model sheds light on the complexity of this recent M 6.5 Walker Lane earthquake.

58 GEOSCIENCES↗
DOE OSTI2021

Scattering amplitudes for all masses and spins

We introduce a formalism for describing four-dimensional scattering amplitudes for particles of any mass and spin. This naturally extends the familiar spinor-helicity formalism for massless particles to one where these variables carry an extra SU(2) little group index for massive particles, with the amplitudes for spin Sparticles transforming as symmetric rank 2S tensors. We systematically characterise all possible three particle amplitudes compatible with Poincare symmetry. Unitarity, in the form of consistent factorization, imposes algebraic conditions that can be used to construct all possible four-particle tree amplitudes. This also gives us a convenient basis in which to expand all possible four-particle amplitudes in terms of what can be called “spinning polynomials”. Many general results of quantum field theory follow the analysis of four-particle scattering, ranging from the set of all possible consistent theories for massless particles, to spin-statistics, and the Weinberg-Witten theorem. We also find a transparent understanding for why massive particles of sufficiently high spin cannot be “elementary”. The Higgs and Super-Higgs mechanisms are naturally discovered as an infrared unification of many disparate helicity amplitudes into a smaller number of massive amplitudes, with a simple understanding for why this can’t be extended to Higgsing for gravitons. We illustrate a number of applications of the formalism at one-loop, giving few-line computations of the electron (g - 2) as well as the beta function and rational terms in QCD. “Off-shell” observables like correlation functions and form-factors can be thought of as scattering amplitudes with external “probe” particles of general mass and spin, so all these objects — amplitudes, form factors and correlators, can be studied from a common on-shell perspective.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2024

Flows in the space of interacting chiral boson theories

We study interacting theories of N left-moving and N ¯ right-moving Floreanini-Jackiw bosons in two dimensions. A parametrized family of such theories is shown to enjoy (nonmanifest) Lorentz invariance if and only if its Lagrangian obeys a flow equation driven by a function of the energy-momentum tensor. We discuss the canonical quantization of such theories along classical stress tensor flows, focusing on the case of the root- T T ¯ deformation, where we obtain perturbative results for the deformed spectrum in a certain large-momentum limit. In the special case N = N ¯ , we consider the quantum effective action for the root- T T ¯ -deformed theory by expanding around a general classical background, and we find that the one-loop contribution vanishes for backgrounds with constant scalar gradients. Our analysis can also be interpreted via dual U ( 1 ) Chern-Simons theories in three dimensions, which might be used to describe deformations of charged AdS 3 black holes or quantum Hall systems. Published by the American Physical Society 2024

Ebert, Stephen (ORCID:0000000154670155)↗
DOE OSTI2021

Situating the Vector Density Approach Among Contemporary Continuum Theories of Dislocation Dynamics

Abstract For the past century, dislocations have been understood to be the carriers of plastic deformation in crystalline solids. However, their collective behavior is still poorly understood. Progress in understanding the collective behavior of dislocations has primarily come in one of two modes: the simulation of systems of interacting discrete dislocations and the treatment of density measures of varying complexity that are considered as continuum fields. A summary of contemporary models of continuum dislocation dynamics is presented. These include, in order of complexity, the two-dimensional statistical theory of dislocations, the field dislocation mechanics treating the total Kröner–Nye tensor, vector density approaches that treat geometrically necessary dislocations on each slip system of a crystal, and high-order theories that examine the effect of dislocation curvature and distribution over orientation. Each of theories contain common themes, including statistical closure of the kinetic dislocation transport equations and treatment of dislocation reactions such as junction formation. An emphasis is placed on how these common themes rely on closure relations obtained by analysis of discrete dislocation dynamics experiments. The outlook of these various continuum theories of dislocation motion is then discussed.

Engineering↗
DOE OSTI2022

Improved Upper Limit on Degree-scale CMB B -mode Polarization Power from the 670 Square-degree POLARBEAR Survey

We report an improved measurement of the degree-scale cosmic microwave background B-mode angular-power spectrum over 670 deg 2 sky area at 150 GHz with Polarbear. In the original analysis of the data, errors in the angle measurement of the continuously rotating half-wave plate, a polarization modulator, caused significant data loss. By introducing an angle-correction algorithm, the data volume is increased by a factor of 1.8. We report a new analysis using the larger data set. We find the measured B-mode spectrum is consistent with the ΛCDM model with Galactic dust foregrounds. We estimate the contamination of the foreground by cross-correlating our data and Planck 143, 217, and 353 GHz measurements, where its spectrum is modeled as a power law in angular scale and a modified blackbody in frequency. We place an upper limit on the tensor-to-scalar ratio r < 0.33 at 95% confidence level after marginalizing over the foreground parameters.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2025

On the discretization error of the discrete generalized quantum master equation

The transfer tensor method (TTM) [Cerrillo and Cao, Phys. Rev. Lett. 112 , 110401 (2014)] can be considered a discrete-time formulation of the Nakajima–Zwanzig quantum master equation (NZ-QME) for modeling non-Markovian quantum dynamics. A recent paper [Makri, J. Chem. Theory Comput. 21 , 5037 (2025)] raised concerns regarding the consistency of the TTM discretization, particularly a spurious term at the initial time t = 0. Here, this work presents a detailed analysis of the discretization structure of the TTM, clarifying the origin of the initial-time correction and establishing a consistent relationship between the TTM discrete-time memory kernel K N and the continuous-time NZ-QME kernel $\mathscr{K}$( N Δ t ). This relationship is validated numerically using the spin-boson model, demonstrating convergence of reconstructed memory kernels and accurate dynamical evolution as Δ t → 0. While the TTM provides a consistent discretization, we note that alternative schemes are also viable, such as the midpoint derivative/midpoint integral scheme proposed in Makri’s work. The relative performance of various schemes for either computing accurate $\mathscr{K}$( N Δ t ) from exact dynamics or obtaining accurate dynamics from exact $\mathscr{K}$( N Δ t ) warrants further investigation.

Density-matrix↗
DOE OSTI2022

In-flight polarization angle calibration for LiteBIRD: blind challenge and cosmological implications

We present a demonstration of the in-flight polarization angle calibration for the JAXA/ISAS second strategic large class mission, LiteBIRD, and estimate its impact on the measurement of the tensor-to-scalar ratio parameter, r, using simulated data. We generate a set of simulated sky maps with CMB and polarized foreground emission, and inject instrumental noise and polarization angle offsets to the 22 (partially overlapping) LiteBIRD frequency channels. Our in-flight angle calibration relies on nulling the EB cross correlation of the polarized signal in each channel. This calibration step has been carried out by two independent groups with a blind analysis, allowing an accuracy of the order of a few arc-minutes to be reached on the estimate of the angle offsets. Both the corrected and uncorrected multi-frequency maps are propagated through the foreground cleaning step, with the goal of computing clean CMB maps. We employ two component separation algorithms, the Bayesian-Separation of Components and Residuals Estimate Tool (B-SeCRET), and the Needlet Internal Linear Combination (NILC). We find that the recovered CMB maps obtained with algorithms that do not make any assumptions about the foreground properties, such as NILC, are only mildly affected by the angle miscalibration. However, polarization angle offsets strongly bias results obtained with the parametric fitting method. Once the miscalibration angles are corrected by EB nulling prior to the component separation, both component separation algorithms result in an unbiased estimation of the r parameter. While this work is motivated by the conceptual design study for LiteBIRD, its framework can be broadly applied to any CMB polarization experiment. In particular, the combination of simulation plus blind analysis provides a robust forecast by taking into account not only detector sensitivity but also systematic effects.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Proxy-based Bayesian inversion of strain tensor data measured during well tests

Recent instrument developments have made it possible to measure the strain tensor caused by injecting or pumping fluid from aquifers or reservoirs, but the full value of these data is limited because the long runtimes of poroelastic forward models makes it impractical to use many inversion schemes. This limits the interpretation of strain data for managing the recovery of resources or storage of wastes in the subsurface. This paper describes a method of inverting deformation data using a poroelastic numerical simulator so the results can be used to manage reservoirs or aquifers. We developed a workflow designed to reduce the number of simulations sufficiently to make it feasible to use DREAMzs, an advanced Bayesian inversion method that translates the uncertainties from different sources into unbiased posterior parameter distributions and uncertainty envelopes around the field data. Using a KNN proxy model for the poroelastic simulator is key to reducing the overall computations, and the workflow includes a strategy for ensuring the proxy model results converge on the results from the simulator. The workflow is tested using an idealized example that verifies the ability to correctly identify parameters and characterize noise used to perturb the data. Field data from an injection test at an oil reservoir near Tulsa, Oklahoma, are also used to evaluate the efficacy of the workflow with a real dataset. The workflow identified 265 history matching solutions out of 1240 total simulation runs (21% acceptance ratio), where the results were used to characterize posterior parameter distribution and evaluate the prediction uncertainty. Furthermore, this workflow is significant because it enables strain tensor, or other geomechanical measurements to be interpreted to guide decision-making during energy and environmental processes in the subsurface.

42 ENGINEERING↗
DOE OSTI2021

GFCCLib: Scalable and efficient coupled-cluster Green's function library for accurately tackling many-body electronic structure problems

Coupled-cluster Green’s function (GFCC) calculation has drawn much attention in the recent years for targeting the molecular and material electronic structure problems from a many-body perspective in a systematically improvable way. However, GFCC calculations on scientific computing clusters usually suffer from expensive higher di- mensional tensor contractions in the complex space, expensive inter-process communi- cation, and severe load imbalance, which limits it’s routine use for tackling electronic structure problems. Here we present a numerical library prototype that is specifically designed for large-scale GFCC calculations. The design of the library is focused on a systematically optimal computing strategy to improve its scalability and efficiency. The performance of the library is demonstrated by the relevant profiling analysis of running GFCC calculations on remote giant computing clusters. The capability of the library is highlighted by computing a wide near valence band of a fullerene C60 molecule for the first time at the GFCCSD level that shows excellent agreement with the experimental spectrum.

Peng, Bo↗
DOE OSTI2020

Research Introduction [Slides]

In the topic of Reverse Time Imaging, we proposed a new IC to reduce computation cost but reserve image resolution for distributed sensor networks. For Induced Seismicity in Oklahoma, we analyzed fault stress state analysis at state scale, and we applied machine learning techniques to polarity picking and seismicity rate forecasting. The results provide better understanding of fault properties, stress field, and the relationships among fault, stress state, injections, and potential seismic hazards. Lastly, for Microseismic Monitoring, we detected and located 770 low-frequency events (5-50 Hz): (1) Shallow events are highly clustered, consistent with the pathway from injection well 13-10A to monitoring well; moment tensor inversion shows dominant tensile cracking; (2) Deep events are scattered and show migration pattern to the basement; moment tensors show that most events are shear cracks.

58 GEOSCIENCES↗
DOE OSTI2022

Hidden conformal symmetries from Killing towers with an application to large-D/CFT

We generalize the notion of hidden conformal symmetry in Kerr/CFT to Kerr-(A)dS black holes in arbitrary dimensions. We build the SL(2, R) generators directly from the Killing tower, whose Killing tensors and Killing vectors enforce the separability of the equations of motion. Our construction amounts to an explicit relationship between hidden conformal symmetries and Killing tensors: we use the Killing tower to build a novel tensor equation connecting the SL(2,R) Casimir with the radial Klein-Gordon operator. For asymptotically flat black holes in four and five dimensions we recover previously known results that were obtained using the ``near-region'' limit and the monodromy method. We then perform a monodromy evaluation of the Klein-Gordon scalar wave equation for all Kerr-(A)dS black holes, finding explicit forms for the zero mode symmetry generators. We also extend this analysis to the large-dimensional Schwarzschild black hole as a step towards buliding a Large-D/CFT correspondence.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

Optimizing Distributed Training on Frontier for Large Language Models

Large language models (LLMs) have demonstrated remarkable success as foundational models, benefiting various downstream applications through fine-tuning. Loss scaling studies have demonstrated the superior performance of larger LLMs compared to their smaller counterparts. Nevertheless, training LLMs with billions of parameters poses significant challenges and requires considerable computational resources. For example, training a one trillion parameter GPT-style model on 20 trillion tokens requires a staggering 120 million exaflops. This research explores efficient distributed training strategies to extract this computation from Frontier, the world's first exascale supercomputer. We enable and investigate various model and data parallel training techniques, such as tensor parallelism, pipeline parallelism, and sharded data parallelism, to facilitate training a trillion-parameter model on Frontier. We empirically assess these techniques and their associated parameters to determine their impact on memory footprint, communication latency, and GPU's computational efficiency. We analyze the complex interplay among these techniques and find a strategy to combine them to achieve high throughput through hyperparameter tuning. We have identified efficient strategies for training large LLMs of varying sizes through empirical analysis and hyperparameter tuning. For 22 Billion, 175 Billion, and 1 Trillion parameters, we achieved GPU throughputs of 38.38%, 36.14%, and 31.96%, respectively. For the training of the 175 Billion parameter model and the 1 Trillion parameter model, we achieved 100% weak scaling efficiency on 1024 and 3072 Mi250X GPUs, respectively. We also achieved strong scaling efficiencies of 89% and 87% for these two models. We trained these models only tens of iterations instead of training till completion.

Yin, Junqi↗