Search NASA⌕ Search

SEARCH · Search NASA

Results for “gradient methods”

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 487 records · Page 27

MRI-alone radiation therapy planning for prostate cancer: Automatic fiducial marker detection

Purpose: The feasibility of radiation therapy treatment planning using substitute computed tomography (sCT) generated from magnetic resonance images (MRIs) has been demonstrated by a number of research groups. One challenge with an MRI-alone workflow is the accurate identification of intraprostatic gold fiducial markers, which are frequently used for prostate localization prior to each dose delivery fraction. This paper investigates a template-matching approach for the detection of these seeds in MRI. Methods: Two different gradient echo T1 and T2* weighted MRI sequences were acquired from fifteen prostate cancer patients and evaluated for seed detection. For training, seed templates from manual contours were selected in a spectral clustering manifold learning framework. This aids in clustering “similar” gold fiducial markers together. The marker with the minimum distance to a cluster centroid was selected as the representative template of that cluster during training. During testing, Gaussian mixture modeling followed by a Markovian model was used in automatic detection of the probable candidates. The probable candidates were rigidly registered to the templates identified from spectral clustering, and a similarity metric is computed for ranking and detection. Results: A fiducial detection accuracy of 95% was obtained compared to manual observations. Expert radiation therapist observers were able to correctly identify all three implanted seeds on 11 of the 15 scans (the proposed method correctly identified all seeds on 10 of the 15). Conclusions: An novel automatic framework for gold fiducial marker detection in MRI is proposed and evaluated with detection accuracies comparable to manual detection. When radiation therapists are unable to determine the seed location in MRI, they refer back to the planning CT (only available in the existing clinical framework); similarly, an automatic quality control is built into the automatic software to ensure that all gold seeds are either correctly detected or a warning is raised for further manual intervention.

60 APPLIED LIFE SCIENCES↗

Evaluation of internal potential distribution and carrier extraction properties of organic solar cells through Kelvin probe and time-of-flight measurements

The carrier extraction property of a prototypical small molecule organic solar cell (OSC) composed of copper phthalocyanine (CuPc), C⁶⁰, and bathocuproine (BCP) was studied on the basis of the internal potential distribution and carrier dynamics in the device. The internal potential distribution in the OSC structure at the interfaces and in the bulk region was determined by the Kelvin probe method. Significant potential gradients were found in the CuPc film on indium tin oxide and in the C⁶⁰ film on CuPc, consistent with charge transfer through the contacts. Moreover, surface potential of the BCP layer grew linearly with increasing film thickness with a slope of ca. 35 mV/nm (giant surface potential: GSP), which indicated spontaneous orientation polarization in the film. The potential gradient and GSP significantly changed the built-in potential of the device. Current–voltage and modified time-of-flight measurements revealed that the BCP layer worked as an electron injection and extraction layer despite the wide energy gap. These results were discussed based on the contributions of GSP and the gap states in the BCP layer.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Assessing the relative importance of flame regimes in Raman/Rayleigh line measurements of turbulent lifted flames

Understanding and quantifying the relative importance of premixed and non-premixed reaction zones within turbulent partially premixed flames is an important issue for multi-regime combustion. The recently-developed method of gradient-free regime identification (GFRI) is applied to instantaneous 1D Raman/Rayleigh measurements of temperature and major species from two turbulent lifted methane/air flames. Local premixed and non-premixed reaction zones are identified using criteria based on the mixture fraction, the chemical explosive mode, and the heat release rate, the latter two being calculated from an approximation of the full thermochemical state of each measured sample. A chemical mode (CM) zero-crossing is a previously documented marker for a premixed reaction zone. Results from the lifted flames show strong correlations among the mixture fraction at the CM zero-crossing, the magnitude of the change in CM at the zero-crossing, and the local heat release rate at the CM zero-crossing compared to the maximum heat release rate. The trends are confirmed through a comparable analysis of numerical simulations of two laminar triple flames. These newly documented trends are associated with the transition from dominantly premixed flame structures to dominantly non-premixed flames structures. The methods introduced for assessing the relative importance of local premixed and non-premixed reactions zones have potential for application to a broad range of turbulent flames.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Half-metallic ferromagnetism in Cr-doped semiconducting Ge-chalcogenide: Density functional approach

A supercell approach has been used to calculate the electronic and magnetic properties of Cr-doped Ge chalcogenide, Ge{sub 1−x}Cr{sub x}Te (x = 0.25 and 0.125). The calculations have been performed using full potential Linear Augmented Plane Wave (FPLAPW) method within generalized gradient approximation (GGA) as exchange-correlation (XC) potential. The calculated results show that the doping of Cr induces the 100% spin polarization at Fermi level (EF) and showed the robust half metallic ferromagnetism in this compound. Thus, the compound at both dopant concentrations behave as dilute magnetic semiconductor (DMS) showing metallic property in majority and semiconducting for minority spin channels which is best suited for spintronic applications. The total magnetic moments of this compound are mainly due to Cr-d states present at E{sup F} with negligible contribution from electronic states of other atoms.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Energy Efficiency of Distributed Environmental Control Systems

In this report, we present an analytical evaluation of the potential of occupant-regulated distributed environmental control systems (DECS) to enhance individual occupant thermal comfort in an office building with no increase, and possibly even a decrease in annual energy consumption. To this end we developed and applied several analytical models that allowed us to optimize comfort and energy consumption in partitioned office buildings equipped with either conventional central HVAC systems or occupant-regulated DECS. Our approach involved the following interrelated components: 1. Development of a simplified lumped-parameter thermal circuit model to compute the annual energy consumption. This was necessitated by the need to perform tens of thousands of optimization calculations involving different US climatic regions, and different occupant thermal preferences of a population of ~50 office occupants. Yearly transient simulations using TRNSYS, a time-dependent building energy modeling program, were run to determine the robustness of the simplified approach against time-dependent simulations. The simplified model predicts yearly energy consumption within approximately 0.6% of an equivalent transient simulation. Simulations of building energy usage were run for a wide variety of climatic regions and control scenarios, including traditional “one-size-fits-all” (OSFA) control; providing a uniform temperature to the entire building, and occupant-selected “have-it-your-way” (HIYW) control with a thermostat at each workstation. The thermal model shows that, un-optimized, DECS would lead to an increase in building energy consumption between 3-16% compared to the conventional approach depending on the climate regional and personal preferences of building occupants. Variations in building shape had little impact in the relative energy usage. 2. Development of a gradient-based optimization method to minimize energy consumption of DECS while keeping each occupant’s thermal dissatisfaction below a given threshold. The DECS energy usage was calculated using the simplified thermal model. OSFA control; providing a uniform temperature to the entire building, and occupant-selected HIYW control with a thermostat at each workstation were implemented for 3 cities representing 3 different climatic regions and control scenarios. It is shown that optimization allows DECS to deliver a higher level of individual and population thermal comfort while achieving annual energy savings between 14 and 26% compared to OSFA. The optimization model also allowed us to study the influence of the partitions’ thermal resistance and the variability of internal loads at each office. These influences didn’t make significant changes in the optimized energy consumption relative to OSFA. The results show that it is possible to provide thermal comfort for each occupant while saving energy compared to OSFA Furthermore, to simplify the implementation of this approach, a fuzzy logic system has been developed to generalize the overall optimization strategy. Its performance was almost as good as the gradient system. The fuzzy system provided thermal comfort to each occupant and saved energy compared to OSFA. The energy savings of the fuzzy system were not as high as for the gradient-optimized system, but the fuzzy system avoided complete connectivity, and the optimization did not have to be repeated for each population. 3. We employed a detailed CFD model of adjacent occupied cubicles to extend the thermal-circuit model in three significant ways: (a) relax the “office wall” requirement by allowing energy to flow between zones via advection as well as conduction, (b) improve the comfort model to account both for radiation as well as convection heat transfer, and (c) support ventilation systems in which the temperature is stratified, such as in underfloor air distribution systems. Initially, three-dimensional CFD simulations of several cubicle configurations, with an adjoining corridor, were performed both to understand the advection between cubicles and the resulting temperature stratification. These simulations showed that the advective flow between cubicles is very significant and severely limits the occupants’ ability to control the personal micro-environments by simply controlling the temperature of the incoming air. Subsequently, the existing thermal-circuit model was extended to include the phenomena described above. The modifications to the thermal-circuit model, which were incorporated such that the simulation time was only slightly impacted, showed that accounting for room stratification resulting from the use of floor swirl diffusers could lead to 10%-26% reduction in the annual energy consumed for HVAC in non-temperate climates. This trend was evident in both OSFA and HIYW scenarios. However, the ratio of energy usage in the two scenarios was little affected by the enhancements in the thermal model.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Electronic band structure and optical properties of titanium oxyphosphates Li{sub 0.50}Co{sub 0.25}TiO(PO{sub 4}) single crystals: An ab-initio calculations

From the refined atomic positions obtained by Belmal et al. (2004) using X-ray diffraction for Li{sub 0.50}Co{sub 0.25}TiO(PO{sub 4}), we have performed a structural optimization by minimizing the forces acting on the atoms keeping the lattice parameters fixed at the experimental values. With this relaxed (optimized) geometry we have performed a comprehensive theoretical study of electronic properties and dispersion of the linear optical susceptibilities using the full potential linear augmented plane wave (FP-LAPW) method. The generalized gradient approximation (GGA) exchange-correlation potential was applied. In addition, the Engel-Vosko generalized gradient approximation (EVGGA) was used for comparison with GGA because it is known that EVGGA approach yields better band splitting compared to the GGA. We have calculated the band structure, and the total and partial densities of states. The electron charge densities and the bonding properties were analyzed and discussed. The complex dielectric optical susceptibilities were discussed in detail. - Graphical abstract: It is shown that P is tetrahedrally coordinated by four O ions. Highlights: > Comprehensive theoretical study of electronic and optical properties was performed. > Using X-ray diffraction data we have performed a structural optimization. > The electron charge densities and the bonding properties were analyzed and discussed. > Fermi surface was analyzed since it is useful for predicting thermal, magnetic, and optical properties. > The density of states at E{sub F} and the electronic specific heat coefficient were calculated.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

A proximal trust-region method for nonsmooth optimization with inexact function and gradient evaluations

Many applications require minimizing the sum of smooth and nonsmooth functions. For example, basis pursuit denoising problems in data science require minimizing a measure of data misfit plus an $\ell^1$-regularizer. Similar problems arise in the optimal control of partial differential equations (PDEs) when sparsity of the control is desired. Here, we develop a novel trust-region method to minimize the sum of a smooth nonconvex function and a nonsmooth convex function. Our method is unique in that it permits and systematically controls the use of inexact objective function and derivative evaluations. When using a quadratic Taylor model for the trust-region subproblem, our algorithm is an inexact, matrix-free proximal Newton-type method that permits indefinite Hessians. We prove global convergence of our method in Hilbert space and demonstrate its efficacy on three examples from data science and PDE-constrained optimization.

97 MATHEMATICS AND COMPUTING↗

Combinatorial Nitrogen Gradients in Sputtered Thin Films

High-throughput synthesis and characterization methods can significantly accelerate the rate of experimental research. For physical vapor deposition (PVD), these methods include combinatorial sputtering with intentional gradients of metal/metalloid composition, temperature, and thickness across the substrate. However, many other synthesis parameters still remain out of reach for combinatorial methods. Here, we extend combinatorial sputtering parameters to include gradients of gaseous elements in thin films. Specifically, a nitrogen gradient was generated in a thin film sample library by placing two MnTe sputtering sources with different gas flows (Ar and Ar/N 2 ) opposite of one another during the synthesis. The nitrogen content gradient was measured along the sample surface, correlating with the distance from the nitrogen source. The phase, composition, and optoelectronic properties of the resulting thin films change as a function of the nitrogen content. Furthermore, this work shows that gradients of gaseous elements can be generated in thin films synthesized by sputtering, expanding the boundaries of combinatorial science.

14 SOLAR ENERGY↗

Validation for 2D/3D registration II: The comparison of intensity- and gradient-based merit functions using a new gold standard data set

Purpose: A new gold standard data set for validation of 2D/3D registration based on a porcine cadaver head with attached fiducial markers was presented in the first part of this article. The advantage of this new phantom is the large amount of soft tissue, which simulates realistic conditions for registration. This article tests the performance of intensity- and gradient-based algorithms for 2D/3D registration using the new phantom data set. Methods: Intensity-based methods with four merit functions, namely, cross correlation, rank correlation, correlation ratio, and mutual information (MI), and two gradient-based algorithms, the backprojection gradient-based (BGB) registration method and the reconstruction gradient-based (RGB) registration method, were compared. Four volumes consisting of CBCT with two fields of view, 64 slice multidetector CT, and magnetic resonance-T1 weighted images were registered to a pair of kV x-ray images and a pair of MV images. A standardized evaluation methodology was employed. Targets were evenly spread over the volumes and 250 starting positions of the 3D volumes with initial displacements of up to 25 mm from the gold standard position were calculated. After the registration, the displacement from the gold standard was retrieved and the root mean square (RMS), mean, and standard deviation mean target registration errors (mTREs) over 250 registrations were derived. Additionally, the following merit properties were computed: Accuracy, capture range, number of minima, risk of nonconvergence, and distinctiveness of optimum for better comparison of the robustness of each merit. Results: Among the merit functions used for the intensity-based method, MI reached the best accuracy with an RMS mTRE down to 1.30 mm. Furthermore, it was the only merit function that could accurately register the CT to the kV x rays with the presence of tissue deformation. As for the gradient-based methods, BGB and RGB methods achieved subvoxel accuracy (RMS mTRE down to 0.56 and 0.70 mm, respectively). Overall, gradient-based similarity measures were found to be substantially more accurate than intensity-based methods and could cope with soft tissue deformation and enabled also accurate registrations of the MR-T1 volume to the kV x-ray image. Conclusions: In this article, the authors demonstrate the usefulness of a new phantom image data set for the evaluation of 2D/3D registration methods, which featured soft tissue deformation. The author's evaluation shows that gradient-based methods are more accurate than intensity-based methods, especially when soft tissue deformation is present. However, the current nonoptimized implementations make them prohibitively slow for practical applications. On the other hand, the speed of the intensity-based method renders these more suitable for clinical use, while the accuracy is still competitive.

60 APPLIED LIFE SCIENCES↗

Distance‐based reconstruction of protein quaternary structures from inter‐chain contacts

Abstract Predicting the quaternary structure of protein complex is an important problem. Inter‐chain residue‐residue contact prediction can provide useful information to guide the ab initio reconstruction of quaternary structures. However, few methods have been developed to build quaternary structures from predicted inter‐chain contacts. Here, we develop the first method based on gradient descent optimization (GD) to build quaternary structures of protein dimers utilizing inter‐chain contacts as distance restraints. We evaluate GD on several datasets of homodimers and heterodimers using true/predicted contacts and monomer structures as input. GD consistently performs better than both simulated annealing and Markov Chain Monte Carlo simulation. Starting from an arbitrarily quaternary structure randomly initialized from the tertiary structures of protein chains and using true inter‐chain contacts as input, GD can reconstruct high‐quality structural models for homodimers and heterodimers with average TM‐score ranging from 0.92 to 0.99 and average interface root mean square distance from 0.72 Å to 1.64 Å. On a dataset of 115 homodimers, using predicted inter‐chain contacts as restraints, the average TM‐score of the structural models built by GD is 0.76. For 46% of the homodimers, high‐quality structural models with TM‐score ≥ 0.9 are reconstructed from predicted contacts. There is a strong correlation between the quality of the reconstructed models and the precision and recall of predicted contacts. Only a moderate precision or recall of inter‐chain contact prediction is needed to build good structural models for most homodimers. Moreover, GD improves the quality of quaternary structures predicted by AlphaFold2 on a Critical Assessment of Techniques for Protein Structure Prediction–Critical Assessments of Predictions of Interactions dataset.

59 BASIC BIOLOGICAL SCIENCES↗

Quasi parton distributions and the gradient flow

We propose a new approach to determining quasi parton distribution functions (PDFs) from lattice quantum chromodynamics. By incorporating the gradient flow, this method guarantees that the lattice quasi PDFs are finite in the continuum limit and evades the thorny, and as yet unresolved, issue of the renormalization of quasi PDFs on the lattice. In the limit that the flow time is much smaller than the length scale set by the nucleon momentum, the moments of the smeared quasi PDF are proportional to those of the lightfront PDF. Finally, we use this relation to derive evolution equations for the matching kernel that relates the smeared quasi PDF and the light-front PDF.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Eigenfrequency constrained topology optimization of finite strain hyperelastic structures

This paper incorporates hyperelastic materials, nonlinear kinematics, and preloads in eigenfrequency constrained density–based topology optimization. The formulation allows for initial finite deformations and subsequent small harmonic oscillations. The optimization problem is solved by the method of moving asymptotes, and the gradients are calculated using the adjoint method. Both simple and degenerate eigenfrequencies are considered in the sensitivity analysis. A well-posed topology optimization problem is formulated by filtering the volume fraction field. Numerical issues associated with excessive distortion and spurious eigenmodes in void regions are reduced by removing low volume fraction elements. The optimization objective is to maximize stiffness subject to a lower bound on the fundamental eigenfrequency. Numerical examples show that the eigenfrequencies drastically change with the load magnitude, and that the optimization is able to produce designs with the desired fundamental eigenfrequency.

42 ENGINEERING↗

Multi-component Reynolds-averaged Navier–Stokes simulations of Richtmyer–Meshkov instability and mixing induced by reshock at different times

Turbulent mixing generated by shock-driven acceleration of a perturbed interface is simulated using a new multi-component Reynolds-averaged Navier–Stokes (RANS) model closed with a two-equation K–ϵ model. Here, the model is implemented in a hydrodynamics code using a third-order weighted essentially non-oscillatory finite-difference method for the advection terms and a second-order central difference method for the gradients in the source and diffusion terms. In the present reshocked Richtmyer–Meshkov instability and mixing study, an incident shock with Mach number Ma s = 1.20 is generated in air and progresses into a sulfur hexafluoride test section. The time evolution of the predicted mixing layer widths corresponding to six shock tube test section lengths are compared with experimental measurements and three-dimensional multi-mode numerical simulations. The mixing layer widths are also compared with the analytical self-similar power-law solution of the simplified model equations prior to reshock. A set of model coefficients and initial conditions specific to these six experiments is established, for which the widths before and after reshock agree very well with experimental and numerical simulation data. A second set of general coefficients that accommodates a broader range of incident shock Mach numbers, Atwood numbers, and test section lengths is also established by incorporating additional experimental data for Ma s = 1.24, 1.50, and 1.98 with At = 0.67 and Ma s = 1.45 with At = –0.67 and previous RANS modeling. Terms in the budgets of the turbulent kinetic energy and dissipation rate equations are examined to evaluate the relative importance of turbulence production, dissipation and diffusion mechanisms during mixing. Convergence results for the mixing layer widths, mean fields, and turbulent fields under grid refinement are presented for each of the Ma s = 1.20 cases.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Machine Learning Analysis of Impact of Western US Fires on Central US Hailstorms

Fires, including wildfires, harm air quality and essential public services like transportation, communication, and utilities. These fires can also influence atmospheric conditions, including temperature and aerosols, potentially affecting severe convective storms. Here, we investigate the remote impacts of fires in the western United States (WUS) on the occurrence of large hail (size: $\geqslant$ 2.54 cm) in the central US (CUS) over the 20-year period of 2001–20 using the machine learning (ML), Random Forest (RF), and Extreme Gradient Boosting (XGB) methods. The developed RF and XGB models demonstrate high accuracy (> 90%) and F1 scores of up to 0.78 in predicting large hail occurrences when WUS fires and CUS hailstorms coincide, particularly in four states (Wyoming, South Dakota, Nebraska, and Kansas). The key contributing variables identified from both ML models include the meteorological variables in the fire region (temperature and moisture), the westerly wind over the plume transport path, and the fire features (i.e., the maximum fire power and burned area). Importantly, the results confirm a linkage between WUS fires and severe weather in the CUS, corroborating the findings of our previous modeling study conducted on case simulations with a detailed physics model.

54 ENVIRONMENTAL SCIENCES↗

Strong Correlation Effect in Ferrimagnetic Half-Metallic V{sub 2}CoAl and V{sub 2}CoGa Heusler Compounds

We present first-principles calculations based on the full-potential linearized augmented plane wave (FP-LAPW) method to investigate the structural, elastic, electronic, and magnetic properties of the Heusler compounds V{sub 2}CoAl and V{sub 2}CoGa. We explore different magnetic configurations in both regular and inverse Heusler structures: the Cu{sub 2}MnAl-type and Hg{sub 2}CuTi-type structures respectively. Our compounds are found to be ferrimagnets in the inverse Heusler structure. To take into account the effects of the strong correlation among the localized d-states, we perform a calculation using the GGA+U method (the general gradient approximation with the Hubbard on-site Coulomb correction U). Then we show that both compounds exhibit a half-metallic character and have a magnetic moment of 2.00 μ{sub B} per formula unit. This latter statement is coherent with the generalized Slater-Pauling rule (SP). In terms of the band structure calculations, we explain the formation of the indirect energy band gaps in the spin-majority channels. We also use the mean-field approximation (MFA) to discuss the magnetic exchange couplings and predict the Curie temperature for both compounds.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Effect of Fe Doping and O Vacancies on the Magnetic Properties of Rutile TiO{sub 2}

Magnetism sources of Fe doping and O vacancies, which coexist in the presence of rutile TiO{sub 2}, are controversial and require a resolution. The effects of Fe doping and point defect on the magnetism of rutile TiO{sub 2} were studied using geometry optimization and energy calculation based on the first principle generalized gradient approximation + U method (GGA+U) of density functional theory. Fe/Vo ratio is 1:1 in rutile TiO{sub 2}. The next closest distance between Fe and Vo is 0.04082 nm. The formation energy of the doping system is the smallest. This doping system has the highest stability. The coexistence of Fe doping and O vacancy achieved ferromagnetic long-range order. The different ratios of Fe/Vo are 1:1, 1:2, 2:1, and 2:2. The magnetic properties of the doping system are significantly different. At a ratio of 2:1, the magnetic moment is the largest among all doping systems. Moreover, the Curie temperature of the doping system can be higher than room temperature. These findings can obtain dilute magnetic semiconductors with potential application value. At a Fe/V{sub O} ratio of 2:2, the Ti{sub 14}Fe{sub 2}O{sub 30} system produces a half-metallic property of 100% electron spin polarizability. This is the most valuable application for designing and preparing novel spintronic injection source dilute magnetic semiconductors. The magnetic source is mainly mediated by the holes generated by Fe doping and O vacancy complexes, thereby causing the spin polarization double exchange effect on the electron of Ti–3d orbital atoms near the O vacancy, O–2p electron orbital, and Fe–3d electron orbital. This feature is consistent with the following theories: the mean field approximation and the double exchange mechanism theories.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Optimal Wetting Angles in Lattice Boltzmann Simulations of Viscous Fingering

We conduct pore-scale simulations of two-phase flow using the 2D Rothman–Keller colour gradient lattice Boltzmann method to study the effect of wettability on saturation at breakthrough (sweep) when the injected fluid first passes through the right boundary of the model. We performed a suite of 189 simulations in which a “red” fluid is injected at the left side of a 2D porous model that is initially saturated with a “blue” fluid spanning viscosity ratios M=ν r /ν b ∈[0.001,100] and wetting angles θ w ∈[0°,180°]. As expected, at low-viscosity ratios M=ν r /ν b $\ll$1 we observe viscous fingering in which narrow tendrils of the red fluid span the model, and for high-viscosity ratios M$\gg$1, we observe stable displacement. The viscous finger morphology is affected by the wetting angle with a tendency for more rounded fingers when the injected fluid is wetting. However, rather than the expected result of increased saturation with increasing wettability, we observe a complex saturation landscape at breakthrough as a function of viscosity ratio and wetting angle that contains hills and valleys with specific wetting angles at given viscosity ratios that maximize sweep. This unexpected result that sweep does not necessarily increase with wettability has major implications to enhanced oil recovery and suggests that the dynamics of multiphase flow in porous media has a complex relationship with the geometry of the medium and the hydrodynamical parameters.

Engineering↗

Speeding up and reducing memory usage for scientific machine learning via mixed precision

Scientific machine learning (SciML) has emerged as a versatile approach to address complex computational science and engineering problems. Within this field, physics-informed neural networks (PINNs) and deep operator networks (DeepONets) stand out as the leading techniques for solving partial differential equations by incorporating both physical equations and experimental data. However, training PINNs and DeepONets require significant computational resources, including long computational times and large amounts of memory. In search of computational efficiency, training neural networks using half precision (float16) rather than the conventional single (float32) or double (float64) precision has gained substantial interest, given the inherent benefits of reduced computational time and memory consumed. However, we find that float16 cannot be applied to SciML methods, because of gradient divergence at the start of training, weight updates going to zero, and the inability to converge to a local minima. To overcome these limitations, we explore mixed precision, which is an approach that combines the float16 and float32 numerical formats to reduce memory usage and increase computational speed. Our experiments showcase that mixed precision training not only substantially decreases training times and memory demands but also maintains model accuracy. Here, we also reinforce our empirical observations with a theoretical analysis. The research has broad implications for SciML in various computational applications.

97 MATHEMATICS AND COMPUTING↗