SEARCH · Search NASA

Results for “distribution shift”

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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 55 records · Page 3

All sources NASA NTRS DOE OSTI

DOE OSTI2023

Neural Image Compression: Generalization, Robustness, and Spectral Biases

Recent advances in neural image compression (NIC) have resulted in models which are starting to outperform traditional codecs. While this has led to growing excitement about using these methods in real-world applications, the successful adoption of any machine learning system (including NIC) in the wild requires it to generalize (and be robust) to unseen distribution shifts at deployment time. Unfortunately, current research lacks comprehensive datasets and informative tools to evaluate and understand compression performance in real-world settings. To bridge this crucial gap, first, this paper presents a comprehensive benchmark suite to evaluate the out-of-distribution (OOD) performance of image compression methods. Specifically, we design CLIC-C and Kodak-C by introducing 15 common corruptions to popular CLIC and Kodak benchmarks. Next, we propose spectrally inspired introspection tools to gain a deeper understanding of errors introduced by image compression methods as well as their OOD performance. To this end, we carry out a detailed performance comparison of the classical codec with various variants of NIC (e.g., original, variable rate, pruned), revealing intriguing findings that challenge our current understanding of the strengths and limitations of NIC. Finally, we corroborate our empirical findings with theoretical analysis, providing an in-depth view of the OOD performance of NIC. Our benchmarks, spectral introspection tools, and findings provide a crucial bridge to the real-world adoption of NIC. We hope that our work will propel future efforts in designing more robust and generalizable NIC methods.

neural networks, Variational Autoencoder, robustne↗

DOE OSTI2017

Data from: "Declines in low-elevation subalpine tree populations outpace growth in high-elevation populations with warming"

This data package contains model data that were used to support conclusions drawn in “Declines in low-elevation subalpine populations outpace growth in high-elevation populations with warming”, by Conlisk et al. 2017. Experimental data collected at field sites within the Alpine Treeline Warming Experiment (ATWE), and data from long-term observational plots were collected on Niwot Ridge, Colorado, USA and used to formulate models contained within the folder “Model_archive” in the zipped folder “Conlisk_etal_JEcol2017_model_archive12022020.zip”. The contents of this compressed folder are described in the data user's guide attached to this archive.There are two folders within the zipped folder - “EngelmannSpruce” and “LimberPine” - for each of the two species in the paper. Models are stored as text files and .sch files can also be opened as text files. However, please note that all these files are specific to the RAMAS Metapop population modeling software, and you will need the program in order to be able to run these models. There are two separate documents, both named “Conlisk_JofEcology_SI_01262017” within “Model_archive”. One is a Microsoft Word file, and the other is a PDF. The former can be opened with Microsoft Word, and the latter can be opened by Adobe Acrobat Reader, or any software compatible with a PDF.------------------1. Species distribution shifts in response to climate change require that recruitment increase beyond current range boundaries. For trees with long lifespans, the importance of climate-sensitive seedling establishment to the pace of range shifts has not been demonstrated quantitatively. 2. Using spatially explicit, stochastic population models combined with data from long-term forest surveys, we explored whether the climate-sensitivity of recruitment observed in climate manipulation experiments was sufficient to alter populations and elevation ranges of two widely distributed, high-elevation North American conifers. 3. Empirically observed, warming-driven declines in recruitment led to rapid modeled population declines at the low-elevation, “warm edge” of subalpine forest and slow emergence of populations beyond the high-elevation, “cool edge”. Because population declines in the forest occurred much faster than population emergence in the alpine, we observed range contraction for both species. For Engelmann spruce, this contraction was permanent over the modeled time horizon, even in the presence of increased moisture. For limber pine, lower sensitivity to warming may facilitate persistence at low elevations – especially in the presence of increased moisture – and rapid establishment above treeline, and, ultimately, expansion into the alpine. 4. Synthesis. Assuming 21st century warming and no additional moisture, population dynamics in high-elevation forests led to transient range contractions for limber pine and potentially permanent range contractions for Engelmann spruce. Thus, limitations to seedling recruitment with warming can constrain the pace of subalpine tree range shifts.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2021

Effect of Helium Flux on Near-surface Helium Accumulation in Plasma-Exposed Tungsten

We report results of object kinetic Monte Carlo (OKMC) simulations aimed at understanding the effect of helium flux on the near-surface helium accumulation in plasma-facing tungsten, which is initially defect-free and has a W(100) surface orientation. These OKMC simulations are performed at 933 K for fluxes ranging from 10 22 to 4 × 10 25 He/m 2 s, with 100 eV helium atoms impinging on a W(100) surface up to a maximum fluence of 4×10 19 He/m 2 . In the near-surface region, helium clusters interact elastically with the free surface. The interaction is attractive and results in the drift of mobile helium clusters towards the surface as well as increased trap mutation rates. The associated kinetics and energetics of the above-mentioned processes obtained from molecular dynamics simulations are also considered. The OKMC simulations indicate that as the flux decreases, the retention of implanted helium decreases, and its depth distribution shifts to deeper below the surface in initially pristine tungsten. Furthermore, the fraction of retained helium diffusing into the bulk increases as well, so much so that for 10 22 He/m 2 s, almost all of the retained helium diffused into the bulk with minimal/negligible near-surface helium accumulation. At a given flux, with increasing fluence, the fraction of retained helium initially decreases and then starts to increase after reaching a minimum. The occurrence of the retention minimum shifts to higher fluences as the flux decreases. Although the near-surface helium accumulation spreads deeper into the material with decreasing flux and increasing fluence, the spread appears to saturate at depths between 80 and 100 nm. Finally, we present a detailed analysis of the influence of helium flux on the size and depth distribution of total helium and helium bubbles.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

NASA NTRS1973

The luminosity function of quasars and its evolution: A comparison of optically selected quasars and quasars found in radio catalogs

The luminosity function of quasars and its evolution are discussed, based on comparison of available data on optically selected quasars and quasars found in radio catalogs. It is assumed that the red shift of quasars is cosmological and the results are expressed in the framework of the Lambda = 0, Q sub Q = 1 cosmological model. The predictions of various density evolution laws are compared with observations of an optically selected sample of quasars and quasar samples from radio catalogs. The differences between the optical luminosity functions, the red shift distributions and the radio to optical luminosity ratios of optically selected quasars and radio quasars rule out luminosity functions where there is complete absence of correlation between radio and optical luminosities. These differences also imply that Schmidt's (1970) luminosity function, where there exists a statistical correlation between radio and optical luminosities, although may be correct for high red shift objects, disagrees with observation at low red shifts. These differences can be accounted for by postulating existence of two classes (1 and 2) of objects.

Petrosian, V.↗

DOE OSTI2024

Boundary-Aware Adversarial Learning Domain Adaption and Active Learning for Cross-Sensor Building Extraction

The use of convolutional neural networks (CNNs) for building extraction from remote sensing images has been widely studied and many public datasets have been made available for accelerating development of these CNN models. Yet adapting pretrained models at scale in real-world scenarios remains a challenging task. The main barrier is that certain new labels are still needed to compensate for domain shifting between the labeled data and new images that potentially cover new geographic locations or that are from a different sensor. In this article, we propose to add informatively labeled samples from a new image pool under the paradigm of active learning. To select the most useful samples based on model uncertainty, we first tackle the problem of uncalibrated uncertainty estimation due to distribution shifting by adapting feature extractors with boundary-based adversarial learning. Calibrated uncertainty is used as the query criterion in the active learning process, where the most uncertain samples are selected for annotation and included for model retraining. The proposed workflow was tested with three data pairs in which each workflow represents a scenario often encountered in real-world applications, including adapting pretrained models to new images collected with different sensors or to new geographic areas where appearances and types of buildings are very different. Compared to several baselines, including random sampling, temperature scaling (a well-known uncertainty calibration technique), different query strategies, and active domain adaptation methods, the proposed workflow shows that strategically querying a smaller set of samples for labeling achieves comparable or better building extraction performance. The proposed method reduces the number of labeled samples required to achieve sufficient model accuracy, thus significantly reducing hundreds of person-hours for labeled data creation. In addition, we include a few considerations when deploying this workflow in a GPU cluster that can be easily adapted to achieve operational building extraction model retraining.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2012

Undamped electrostatic plasma waves

Electrostatic waves in a collision-free unmagnetized plasma of electrons with fixed ions are investigated for electron equilibrium velocity distribution functions that deviate slightly from Maxwellian. Of interest are undamped waves that are the small amplitude limit of nonlinear excitations, such as electron acoustic waves (EAWs). A deviation consisting of a small plateau, a region with zero velocity derivative over a width that is a very small fraction of the electron thermal speed, is shown to give rise to new undamped modes, which here are named corner modes. The presence of the plateau turns off Landau damping and allows oscillations with phase speeds within the plateau. These undamped waves are obtained in a wide region of the (k,{omega}{sub R}) plane ({omega}{sub R} being the real part of the wave frequency and k the wavenumber), away from the well-known 'thumb curve' for Langmuir waves and EAWs based on the Maxwellian. Results of nonlinear Vlasov-Poisson simulations that corroborate the existence of these modes are described. It is also shown that deviations caused by fattening the tail of the distribution shift roots off of the thumb curve toward lower k-values and chopping the tail shifts them toward higher k-values. In addition, a rule of thumb is obtained for assessing how the existence of a plateau shifts roots off of the thumb curve. Suggestions are made for interpreting experimental observations of electrostatic waves, such as recent ones in nonneutral plasmas.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

DOE OSTI2020

Laser power density dependence on charge state distribution of Ta ion laser plasma

Laser power density per pulse, which is commonly expressed with the unit of “W/cm2,” is an important parameter to characterize ablation plasma. To match a design charge state of heavy ion beam induced by a laser ion source, a laser power density must be carefully chosen. Above around 108 W/cm2 of laser power density, laser ablation plasma is emitted from the surface of solid material. Then, up to 109 W/cm2, the most abundant charge state is 1+. Because the ionization energy increases with higher charge states, increasing the laser intensity leads to the charge state distribution shifting higher. Increasing the density to increase charge states also results in lower time of flight due to higher velocities. The maximum laser power density is obtained by the smallest available laser spot size on the target material which is determined by the quality of the laser beam. For many accelerator applications, higher charge state beams are preferred. In particular cases, singly charge ion beams are demanded. Therefore, production of intermediate charge state beams has not been investigated well. In this study, we selected Ta4+ as an example demanded beam and tried to clarify how the transition of charge state distribution depends on laser power density. Conclusively, the possible specification of a laser ion source for Ta4+ delivery was elucidated.

Okamura, Masahiro (ORCID:0000000255521706)↗

DOE OSTI2014

Communication: Direct comparison between theory and experiment for correlated angular and product-state distributions of the ground-state and stretching-excited O({sup 3}P) + CH{sub 4} reactions

Motivated by a recent experiment [H. Pan and K. Liu, J. Chem. Phys. 140, 191101 (2014)], we report a quasiclassical trajectory study of the O({sup 3}P) + CH{sub 4}(v{sub k} = 0, 1) → OH + CH{sub 3} [k = 1 and 3] reactions on an ab initio potential energy surface. The computed angular distributions and cross sections correlated to the OH(v = 0, 1) + CH{sub 3}(v = 0) coincident product states can be directly compared to experiment for O + CH{sub 4}(v{sub 3} = 0, 1). Both theory and experiment show that the ground-state reaction is backward scattered, whereas the angular distributions shift toward sideways and forward directions upon antisymmetric stretching (v{sub 3}) excitation of the reactant. Theory predicts similar behavior for the O + CH{sub 4}(v{sub 1} = 1) reaction. The simulations show that stretching excitation enhances the reaction up to about 15 kcal/mol collision energy, whereas the O + CH{sub 4}(v{sub k} = 1) reactions produce smaller cross sections for OH(v = 1) + CH{sub 3}(v = 0) than those of O + CH{sub 4}(v = 0) → OH(v = 0) + CH{sub 3}(v = 0). The former finding agrees with experiment and the latter awaits for confirmation. The computed cold OH rotational distributions of O + CH{sub 4}(v = 0) are in good agreement with experiment.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2023

Adaptive autoencoder latent space tuning for more robust machine learning beyond the training set for six-dimensional phase space diagnostics of a time-varying ultrafast electron-diffraction compact accelerator

In this work, we present a general adaptive latent space tuning approach for improving the robustness of machine learning tools with respect to time variation and distribution shift. We demonstrate our approach by developing an encoder-decoder convolutional neural network-based virtual 6D phase space diagnostic of charged particle beams in the HiRES ultrafast electron diffraction (UED) compact particle accelerator with uncertainty quantification. Our method utilizes model-independent adaptive feedback to tune a low dimensional 2D latent space representation of ~1 million dimensional objects which are the 15 unique 2D projections (x, y),...,(z, p z ) of the 6D phase space (x, y, z, p x , p y , p z ) of the charged particle beams. We demonstrate our method with numerical studies of short electron bunches utilizing experimentally measured UED input beam distributions.

43 PARTICLE ACCELERATORS↗

DOE OSTI2015

The theory behind, and the challenges of, conserving nature's stage in a time of rapid change

Most conservation planning to date has focused on protecting today's biodiversity with the assumption that it will be tomorrow's biodiversity. However, modern climate change has already resulted in distributional shifts of some species and is projected to result in many more shifts in the coming decades. As species redistribute and biotic communities reorganize, conservation plans based on current patterns of biodiversity may fail to adequately protect species in the future. One approach for addressing this issue is to focus on conserving a range of abiotic conditions in the conservation-planning process. By doing so, it may be possible to conserve an abiotically diverse "stage" upon which evolution will play out and support many actors (biodiversity). We reviewed the fundamental underpinnings of the concept of conserving the abiotic stage, starting with the early observations of von Humboldt, who mapped the concordance of abiotic conditions and vegetation, and progressing to the concept of the ecological niche. We discuss challenges posed by issues of spatial and temporal scale, the role of biotic drivers of species distributions, and latitudinal and topographic variation in relationships between climate and landform. For example, abiotic conditions are not static, but change through time-albeit at different and often relatively slow rates. In some places, biotic interactions play a substantial role in structuring patterns of biodiversity, meaning that patterns of biodiversity may be less tightly linked to the abiotic stage. Furthermore, abiotic drivers of biodiversity can change with latitude and topographic position, meaning that the abiotic stage may need to be defined differently in different places. Lastly, we conclude that protecting a diversity of abiotic conditions will likely best conserve biodiversity into the future in places where abiotic drivers of species distributions are strong relative to biotic drivers, where the diversity of abiotic settings will be conserved through time, and where connectivity allows for movement among areas providing different abiotic conditions.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2016

THE HUBBLE SPACE TELESCOPE UV LEGACY SURVEY OF GALACTIC GLOBULAR CLUSTERS. VII. IMPLICATIONS FROM THE NEARLY UNIVERSAL NATURE OF HORIZONTAL BRANCH DISCONTINUITIES

The UV-initiative Hubble Space Telescope Treasury survey of Galactic globular clusters provides a new window into the phenomena that shape the morphological features of the horizontal branch (HB). Using this large and homogeneous catalog of UV and blue photometry, we demonstrate that the HB exhibits discontinuities that are remarkably consistent in color (effective temperature). This consistency is apparent even among some of the most massive clusters hosting multiple distinct sub-populations (such as NGC 2808, ω Cen, and NGC 6715), demonstrating that these phenomena are primarily driven by atmospheric physics that is independent of the underlying population properties. However, inconsistencies arise in the metal-rich clusters NGC 6388 and NGC 6441, where the discontinuity within the blue HB (BHB) distribution shifts ∼1000–2000 K hotter. We demonstrate that this shift is likely due to a large helium enhancement in the BHB stars of these clusters, which in turn affects the surface convection and evolution of such stars. Our survey also increases the number of Galactic globular clusters known to host blue-hook stars (also known as late hot flashers) from 6 to 23 clusters. These clusters are biased toward the bright end of the globular cluster luminosity function, confirming that blue-hook stars tend to form in the most massive clusters with significant self-enrichment.

79 ASTRONOMY AND ASTROPHYSICS↗

DOE OSTI2022

Assurance by Design for Cyber Physical Data-Driven Systems

Currently, Cyber Physical Data-Driven Systems (CPDDS) employ machine learning for the classification, data fusion, and control of our nation’s infrastructure, such as the power grid, transportation networks (e.g., fuel distribution, air traffic control), and DoD long-duration collaborative autonomous platforms including unmanned underwater, ground, surface, space, and aerial systems. Many CPDDSs are system-of-systems that should be designed to communicate over disadvantaged networks. It is important to assure that the CPDDSs are resilient against physical and cyber threats by design. Additionally, their design should tolerate misclassification errors resulting from natural and/or adversarial distribution shifts within their data driven components. The all-domain nature of the problem of assuring the design of CPDDSs requires a multi-disciplinary perspective as outlined in this chapter.

Chikkagoudar, Satish↗

DOE OSTI2020

MimicGAN: Robust Projection onto Image Manifolds with Corruption Mimicking

In the past few years, Generative Adversarial Networks (GANs) have dramatically advanced our ability to represent and parameterize high-dimensional, non-linear image manifolds. As a result, they have been widely adopted across a variety of applications, ranging from challenging inverse problems like image completion, to problems such as anomaly detection and adversarial defense. A recurring theme in many of these applications is the notion of projecting an image observation onto the manifold that is inferred by the generator. In this context, Projected Gradient Descent (PGD) has been the most popular approach, which essentially optimizes for a latent vector that minimizes the discrepancy between a generated image and the given observation. However, PGD is a brittle optimization technique that fails to identify the right projection (or latent vector) when the observation is corrupted, or perturbed even by a small amount. Such corruptions are common in the real world, for example images in the wild come with unknown crops, rotations, missing pixels, or other kinds of non-linear distributional shifts which break current encoding methods, rendering downstream applications unusable. To address this, we propose corruption mimicking—a new robust projection technique, that utilizes a surrogate network to approximate the unknown corruption directly at test time, without the need for additional supervision or data augmentation. The proposed method is significantly more robust than PGD and other competing methods under a wide variety of corruptions, thereby enabling a more effective use of GANs in real-world applications. Finally, more importantly, we show that our approach produces state-of-the-art performance in several GAN-based applications—anomaly detection, domain adaptation, and adversarial defense, that benefit from an accurate projection.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2015

Characterisation of radiation damage in W and W-based alloys from 2MeV self-ion near-bulk implantations

The displacement damage induced in bulk W and W-5 wt.% Re and W-5 wt.% Ta alloys by 2 MeV W + irradiation to doses 3.3×10 17 - 2.5×10 19 W + /m 2 at temperatures ranging from 300 to750°C has been characterized by transmission electron microscopy. An automated sizing and counting approach based on Image J has been proposed and performed for all irradiation data. In all cases the damage comprised dislocation loops, mostly of interstitial type, with Burgers vectors b = ½<111> (> 60%) and b = <100>. The diameters of loops did not exceed 20 nm, with the majority being ≤ 6 nm. The loop number density varied between 10 22 and 10 23 loops/m 3 . With increasing irradiation temperature, the loop size distributions shifted towards larger sizes, and there was a substantial decrease in loop number densities. The damage microstructure was less sensitive to dose than to temperature. Under the same irradiation conditions, loop number densities in the alloys were higher than in pure W but loops were smaller. In grains with normals close to z = <001>, loop strings developed in W at temperatures ≥ 500°C and doses ≥ 1.2 dpa, but such strings were not observed in the W-Re or W-Ta alloys. However, in other grain orientations complex structures appeared in all materials and dense dislocation networks formed at higher doses.

36 MATERIALS SCIENCE↗

DOE OSTI2022

Hydrogenolysis of n -eicosane over Ru-based catalysts in a continuous flow reactor

The hydrogenolysis of n-eicosane (C 20 H 42 ) was studied under differential conditions over Ru-based catalysts including Ru/SiO 2 , Ru/TiO 2 , and Ru/SBA-15, using a continuous flow reactor that was specially designed for large hydrocarbon reactants that are solids at room temperature. Similar rates and product distributions were obtained for Ru/SiO 2 and Ru/SBA-15 at 433 K, with the most abundant products being C 1 and C 19 hydrocarbons and smaller amounts of C 8 to C 18 also being produced. The higher yield of C 1 and C 19 indicates that cleavage of the terminal C-C bond is somewhat more facile than an internal C-C bond. At 473 K the product distribution shifted to mostly C 1 suggesting that, at this temperature, the adsorbed alkyl fragments undergo complete C-C bond scission prior to desorbing. Here, the conversion was also found to be inversely proportional to the H 2 pressure. Similar results were obtained for Ru/TiO 2 except this catalyst was found to be less active than both Ru/SiO 2 and Ru/SBA-15.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2023

Electrochemically induced fracture in LLZO: How the interplay between flaw density and electrostatic potential affects operability

Fracture and short circuit in the Li 7 La 3 Zr 2 O 12 (LLZO) solid electrolyte are two key issues that prevent its adoption in battery cells. In this paper, we utilize phase-field simulations that couple electrochemistry and fracture to evaluate the maximum electric potential that LLZO electrolytes can support as a function of crack density. In the case of a single crack, we find that the applied potential at the onset of crack propagation exhibits inverse square root scaling with respect to crack length, analogous to classical fracture mechanics. Here, we further find that the short-circuit potential scales linearly with crack length. In the realistic case where the solid electrolyte contains multiple cracks, we reveal that failure fits the Weibull model. The failure distributions shift to favor failure at lower overpotentials as areal crack density increases. Furthermore, when flawless interfacial buffers are placed between the applied potential and the bulk of the electrolyte, failure is mitigated. When constant currents are applied, current focuses in near-surface flaws, leading to crack propagation and short circuit. We find that buffered samples sustain larger currents without reaching unstable overpotentials and without failing. Our findings suggest several mitigation strategies for improving the ability of LLZO to support larger currents and improve operability.

25 ENERGY STORAGE↗

DOE OSTI2025

Object detection with deep learning for rare event search in the GADGET II TPC

In the pursuit of identifying rare two-particle events within the GADGET II Time Projection Chamber (TPC), this paper presents a comprehensive approach for leveraging Convolutional Neural Networks (CNNs) and various data processing methods. To address the inherent complexities of 3D TPC track reconstructions, the data is expressed in 2D projections and 1D quantities. This approach capitalizes on the diverse data modalities of the TPC, allowing for the efficient representation of the distinct features of the 3D events, with no loss in topology uniqueness. Additionally, it leverages the computational efficiency of 2D CNNs and benefits from the extensive availability of pre-trained models. Given the scarcity of real training data for the rare events of interest, simulated events are used to train the models to detect real events. To account for potential distribution shifts when predominantly depending on simulations, significant perturbations are embedded within the simulations. This produces a broad parameter space that works to account for potential physics parameter and detector response variations and uncertainties. These parameter-varied simulations are used to train sensitive 2D CNN object detectors. When combined with 1D histogram peak detection algorithms, this multi-modal detection framework is highly adept at identifying rare, two-particle events in data taken during experiment 21072 at the Facility for Rare Isotope Beams (FRIB), demonstrating a 100% recall for events of interest. Here, we present the methods and outcomes of our investigation and discuss the potential future applications of these techniques.

Convolutional neural network↗

DOE OSTI2021

Entropic Control of HD Exchange Rates over Dilute Pd-in-Au Alloy Nanoparticle Catalysts

Dilute Pd-in-Au alloy catalysts are promising materials for selective hydrogenation catalysis. Extensive surface science studies have contributed mechanistic insight on the energetic aspect of hydrogen dissociation, migration and recombination on dilute alloy systems. Yet, translating these fundamental concepts to the kinetics and free energy of hydrogen dissociation on nanoparticle catalysts operating at ambient pressures and temperatures remains challenging. Here, the effect of the Pd concentration and Pd ensemble size on the catalytic activity, apparent activation energy and rate limiting process is addressed by combining experiment and theory. Experiments in a flow reactor show that a compositional change from 4 to 8 atm% Pd of the Pd-in-Au alloy catalyst leads to strong increase in activity, even exceeding the activity per Pd atom of monometallic Pd under the same conditions, albeit with an increase in apparent activation energy. First-principles calculations show that the rate and apparent activation enthalpy for HD exchange increase when increasing the Pd ensemble size from single Pd atoms to Pd trimers in a Au surface, suggesting that the ensemble size distribution shifts from mainly single Pd atoms on the 4 atm% Pd alloy to larger Pd ensembles of at least three atoms for the 8 atm% Pd/Au catalyst. The DFT studies also indicated that the rate-controlling process is different: H 2 (D 2 ) dissociation determines the rate for single atoms whereas recombination of adsorbed H and D determines the rate on Pd trimers. Both experiment and theory suggest that the increased reaction rate with increasing Pd content and ensemble size stems from an entropic driving force. Lastly, our results support hydrogen migration between Pd sites via Au and indicate that the dilute alloy design prevents the formation of subsurface hydrogen, which is crucial in achieving high selectivity in hydrogenation catalysis.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗