Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Cluster 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 1,063 records · Page 59
All sourcesNASA NTRSDOE OSTI
DOE OSTI2013

Structure, stability, and electronic property of carbon-doped gold clusters Au{sub n}C{sup −} (n = 1–10): A density functional theory study

The equilibrium geometric structures, relative stabilities, and electronic properties of Au{sub n}C{sup −} and Au{sub n+1}{sup −} (n = 1–10) clusters are systematically investigated using density functional theory with hyper-generalized gradient approximation. The optimized geometries show that one Au atom capped on Au{sub n−1}C{sup −} clusters is a dominant growth pattern for Au{sub n}C{sup −} clusters. In contrast to Au{sub n+1}{sup −} clusters, Au{sub n}C{sup −} clusters are most stable in a quasi-planar or three-dimensional structure because C doping induces the local non-planarity while the rest of the structure continues to grow in a planar mode, resulting in an overall non-2D configuration. The relative stability calculations show that the impurity C atom can significantly enhance the thermodynamic stability of pure gold clusters. Moreover, the effect of C atom on the Au{sub n}{sup −} host decreases with the increase of cluster size. The HOMO-LUMO gap curves show that the interaction of the C atom with Au{sub n}{sup −} clusters improves the chemical stability of pure gold clusters, except for Au{sub 3}{sup −} and Au{sub 4}{sup −} clusters. In addition, a natural population analysis shows that the charges in corresponding Au{sub n}C{sup −} clusters transfer from the Au{sub n}{sup −} host to the C atom. Meanwhile, a natural electronic configuration analysis also shows that the charges mainly transfer between the 2s and 2p orbitals within the C atom.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

Constraints on the normal branch of DGP gravity from SPT galaxy clusters with DES and HST weak-lensing mass calibration and from P l a n c k PR4 CMB anisotropies

We present constraints on the normal branch of the Dvali-Gabadadze-Porrati (nDGP) braneworld gravity model from the abundance of massive galaxy clusters. On scales below the nDGP crossover scale r c , the nDGP model features an effective gravitylike fifth force that alters the growth of structure, leading to an enhancement of the halo mass function (HMF) on cluster scales. The enhanced cluster abundance allows for constraints on the nDGP model using cluster samples. We employ the South Pole Telescope (SPT) cluster sample, selected through the thermal Sunyaev-Zel’dovich effect with the SPT and with mass calibration using weak-lensing data from the Dark Energy Survey (DES) and the Hubble Space Telescope (HST). The cluster sample contains 1,005 clusters with redshifts 0.25 < z < 1.78 , which are confirmed with the multicomponent matched filter algorithm using optical and near-infrared data. Weak-lensing data from DES and HST enable a robust mass measurement of the cluster sample. We use DES Year 3 data for 688 clusters with redshifts z < 0.95 , and HST data for 39 clusters with redshifts 0.6 < z < 1.7 . We account for the enhancement in the HMF through a semi-analytic correction factor to the standard cosmology HMF derived from the spherical collapse model in the nDGP model. We then further calibrate this model using N -body simulations. In addition, for the first time, we analyze the primary CMB temperature and polarization anisotropy measurements from Planck PR4 within the nDGP model. We obtain a competitive constraint from the joint analysis of the SPT cluster abundance with the Planck PR4 data, and report an upper bound of 1 / H 0 r c < 1.41 at 95% when assuming a cosmology with massive neutrinos.

Vogt, S. M.L. [Munich U. Observ.; LMU Munich (main↗
NASA NTRS2013

Hadoop for High-Performance Climate Analytics: Use Cases and Lessons Learned

Scientific data services are a critical aspect of the NASA Center for Climate Simulations mission (NCCS). Hadoop, via MapReduce, provides an approach to high-performance analytics that is proving to be useful to data intensive problems in climate research. It offers an analysis paradigm that uses clusters of computers and combines distributed storage of large data sets with parallel computation. The NCCS is particularly interested in the potential of Hadoop to speed up basic operations common to a wide range of analyses. In order to evaluate this potential, we prototyped a series of canonical MapReduce operations over a test suite of observational and climate simulation datasets. The initial focus was on averaging operations over arbitrary spatial and temporal extents within Modern Era Retrospective- Analysis for Research and Applications (MERRA) data. After preliminary results suggested that this approach improves efficiencies within data intensive analytic workflows, we invested in building a cyber infrastructure resource for developing a new generation of climate data analysis capabilities using Hadoop. This resource is focused on reducing the time spent in the preparation of reanalysis data used in data-model inter-comparison, a long sought goal of the climate community. This paper summarizes the related use cases and lessons learned.

analytics↗
NASA NTRS1987

HEAO 1 hard X-ray observations of three Abell clusters of galaxies

The results of an analysis of the HEAO 1 hard X-ray measurements of the Coma, A1367, and A2319 galactic clusters are presented. All three clusters have been observed to have diffuse, extended radio emission in their central regions. The results are analyzed together with radio measurements to determine the energetic electron density and the strength of the mean intracluster magnetic field in these clusters. The results show that there is no clear evidence for nonthermal flux in the spectra of the three clusters at the sensitivity limit of the low energy detectors used.

Rephaeli, Y.↗
NASA NTRS2002

A Chandra X-Ray Study of Cygnus A: The Cluster of Galaxies - 3

We present an analysis of the Chandra Advanced CCD Imaging Spectrometer (ACIS) observation of the intracluster gas associated with the cluster of galaxies surrounding Cygnus A. The dominant gaseous structure is a roughly elliptical (presumably prolate spheroidal in three dimensions) feature with semimajor axis approximately equal to 1.(prime)1 (approximately equal to 100 kpc). This structure apparently represents intracluster gas that has been swept up and compressed by a cavity inflated in this gas by relativistic material that has passed through the ends of the radio jets. The X-ray-emitting gas shows this prolate spheroidal morphology to approximately equal to 1.(prime)2 (110 kpc) from the radio galaxy but is spherical on larger scales. The X-ray emission from the intracluster gas extends to at least 8 prime (approximately equal to 720 kpc) from the radio galaxy, and a second, extended source of X-ray emission (probably associated with a second cluster of galaxies) is seen some 12 prime (approximately equal to 1 Mpc) to the northwest of Cygnus A. The X-ray spectrum of the integrated intracluster gas imaged on the S3 chip (dimensions 8 prime x 8 prime = 720 x 720 kpc), excluding the contribution from the radio galaxy and other compact sources of X-ray emission, has a gas temperature, metallicity, and unabsorbed 2-10 keV rest-frame luminosity of 7.7 keV, 0.34 times solar, and 3.5 x 10(exp 44) ergs per second, respectively. We have projected the X-ray spectra taken from 12 elliptical and circular annuli in order to derive a run of temperature, metallicity, density, and pressure as a function of radius. The temperature of the X-ray-emitting gas drops from approximately equal to 8 keV more than 100 kpc from the center to approximately equal to 5 keV some 80 kpc from the center, with the coolest gas immediately adjacent to the radio galaxy. "Belts" of slightly cooler (approximately equal to 4 keV) X-ray-emitting gas run around the minor dimension of the cavity created by the radio source, while the limb-brightened edges of the cavity are slightly hotter (approximately equal to 6 keV), perhaps as a result of heating by a bow shock driven by the probably expanding cavity into the intracluster gas. There is a metallicity gradient in the X-ray-emitting gas, with the highest metallicities (approximately solar) found close to the center, decreasing to approximately 0.3 times solar in the outer parts. We have used the assumption of hydrostatic equilibrium to derive a total cluster mass within 500 kpc of 2.0 x 10(exp 14) solar mass and 2.8 x 10(exp 14) solar mass for constant and centrally decreasing temperature profiles, respectively. The total mass of X-ray-emitting gas within the same radius is 1.1 x 10(exp 13) solar mass. Thus, the gas fraction of the cluster within 500 kpc is 0.055 and 0.039 for the constant and centrally decreasing temperature profiles, respectively.

Smith, David A.↗
NASA NTRS2012

Preliminary Evaluation of MapReduce for High-Performance Climate Data Analysis

MapReduce is an approach to high-performance analytics that may be useful to data intensive problems in climate research. It offers an analysis paradigm that uses clusters of computers and combines distributed storage of large data sets with parallel computation. We are particularly interested in the potential of MapReduce to speed up basic operations common to a wide range of analyses. In order to evaluate this potential, we are prototyping a series of canonical MapReduce operations over a test suite of observational and climate simulation datasets. Our initial focus has been on averaging operations over arbitrary spatial and temporal extents within Modern Era Retrospective- Analysis for Research and Applications (MERRA) data. Preliminary results suggest this approach can improve efficiencies within data intensive analytic workflows.

Duffy, Daniel Q.↗
DOE OSTI2021

Signatures of large-scale cold fronts in the optically-selected merging cluster HSC J085024+001536

Abstract We represent a joint X-ray, weak-lensing, and optical analysis of the optically-selected merging cluster, HSC J085024+001536, from the Subaru HSC-SSP survey. Both the member galaxy density and the weak-lensing mass map show that the cluster is composed of south-east and north-west components. The two-dimensional weak-lensing analysis shows that the south-east component is the main cluster, and the sub-cluster and main cluster mass ratio is $0.32^{+0.75}_{-0.23}$. The north-west sub-cluster is offset by ∼700 kpc from the main cluster center, and their relative line-of-sight velocity is ∼1300 km s−1 from spectroscopic redshifts of member galaxies. The X-ray emission is concentrated around the main cluster, while the gas mass fraction within a sphere of 1′ radius of the sub-cluster is only $f_{\mathrm{gas}}=4.0^{+2.3}_{-3.3}\%$, indicating that the sub-cluster gas was stripped by ram pressure. An X-ray residual image shows three arc-like excess patterns, of which two are symmetrically located at ∼550 kpc from the X-ray morphological center, and the other is close to the X-ray core. The excess close to the sub-cluster has a cold-front feature where dense-cold gas and thin-hot gas contact. The two outer excesses are tangentially elongated about ∼450–650 kpc, suggesting that the cluster is merged with a non-zero impact parameter. Overall features revealed by the multi-wavelength datasets indicate that the cluster is at the second impact or later. Since the optically-defined merger catalog is unbiased for merger boost of the intracluster medium, X-ray follow-up observations will pave the way to understand merger physics at various phases.

Tanaka, Keigo↗
DOE OSTI2022

Microhydration of the metastable N -Protomer of 4-Aminobenzoic acid by condensation at 80 K: H/D exchange without conversion to the more stable O- protomer

4-Aminobenzoic acid (4ABA) is a model scaffold for studying solvent-mediated proton transfer. Although protonation at the carboxylic group (O-protomer) is energetically favored in the gas phase, the N-protomer, where the proton remains on the amino group, can be kinetically trapped by electrospray ionization of 4ABA in an aprotic solvent such as acetonitrile. Here we report the formation of the hydrated deuterium isotopologues of the N-protomers, RND 3 + ·(H 2 O) n=1-3 , (R=C 6 H 4 COOD), which are generated by condensing water molecules onto the bare N-protomers in a liquid nitrogen cooled, radiofrequency octopole ion trap at 80 K. The product clusters are then transferred to a 20 K cryogenic ion trap where they are tagged with weakly bound D 2 molecules. The structures of these clusters are determined by analysis of their vibrational patterns obtained by resonant IR photodissociation. The resulting patterns confirm that the metastable N-protomer configuration remains intact even when warmed by sequential condensation of water molecules. Attachment of H 2 O molecules onto the RND 3 + head group also affords the opportunity to explore the possibility of H/D exchange between the acid scaffold and the proximal water network. The spectroscopic results establish that although the RND 3 + ·(H 2 O) n=1,2 clusters are formed without H/D exchange, the n = 3 cluster exhibits about 10% H/D exchange as evidenced by the appearance of the telltale HOD bands. Furthermore, the site of exchange on the acid is determined to be the acidic OH by the emergence of the OH stretching fundamental in the -COOH motif.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2018

Distribution of defect clusters in the primary damage of ion irradiated 3C-SiC

We report a statistical analysis of sizes and compositions of clusters produced in cascades during irradiation of SiC. The results are obtained using molecular dynamics (MD) simulations of cascades caused by primary knock-on atoms (PKAs) with energies between 10 eV and 50 keV. The results are averaged over six crystallographic directions of the PKA and integrated over PKA energy spectrum derived from the Stopping and Range of Ions in Matter (SRIM) code. Specific results are presented for 1 MeV Kr ion as an example of an impinging particle. We find that distributions of cluster size n for both vacancies and interstitials obey a power law f = A/n S and these clusters are dominated by carbons defects. The fitted values of A and S are different than values reported for metals, which can be explained through different defect energetics and cascade morphology between the two classes of materials. In SiC, the average carbon ratio for interstitial clusters is 91.5%, which is higher than the ratio of C in vacancy clusters, which is 85.3%. Size and composition distribution of in-cascade clusters provide a critical input for long-term defect evolution models.

36 MATERIALS SCIENCE↗
DOE OSTI2019

A Detailed Study of the Most Relaxed SPT-selected Galaxy Clusters: Properties of the Cool Core and Central Galaxy

We present a multi-wavelength analysis of the four most relaxed clusters in the South Pole Telescope 2500 deg 2 survey, which lie at 0.55 < z < 0.75. This study, which utilizes new, deep data from Chandra and Hubble, along with ground-based spectroscopy from Gemini and Magellan, improves significantly on previous studies in both depth and angular resolution, allowing us to directly compare to clusters at z~0. We find that the temperature, density, and entropy profiles of the intracluster medium (ICM) are very similar among the four clusters, and share similar shapes to clusters at z~0. Specifically, we find no evidence for deviations from self similarity in the temperature profile over the radial range 10kpc < r < 1Mpc, implying that the processes responsible for preventing runaway cooling over the past >6 Gyr are, at least roughly, preserving self similarity. We find typical metallicities of ~0.3 Zsun in the bulk of the ICM, rising to ~0.5 Zsun in the inner ~100 kpc, and reaching ~1 Zsun at r < 10kpc. This central excess is similar in magnitude to what is observed in the most relaxed clusters at z~0, suggesting that both the global metallicity and the central excess that we see in cool core clusters at z~0 were in place very early in the cluster lifetime and, specifically, that the central excess is not due to late-time enrichment by the central galaxy. Consistent with observations at z~0, we measure a diversity of stellar populations in the central brightest cluster galaxies of these four clusters, with star formation rates spanning a factor of ~500, despite the similarity in cooling time, cooling rate, and central entropy. These data suggest that, while the details vary dramatically from system to system, runaway cooling has been broadly regulated in relaxed clusters over the past 6 Gyr.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2015

Fast Fault Injection and Sensitivity Analysis for Collective Communications, In: 2015 IEEE International Conference on Cluster Computing

The collective communication operations, which are widely used in parallel applications for global communication and synchronization are critical for application's performance and scalability. However, how faulty collective communications impact the application and how errors propagate between the application processes is largely unexplored. One of the critical reasons for this situation is the lack of fast evaluation method to investigate the impacts of faulty collective operations. The traditional random fault injection methods relying on a large amount of fault injection tests to ensure statistical significance require a significant amount of resources and time. These methods result in prohibitive evaluation cost when applied to the collectives. In this paper, we introduce a novel tool named Fast Fault Injection and Sensitivity Analysis Tool (FastFIT) to conduct fast fault injection and characterize the application sensitivity to faulty collectives. The tool achieves fast exploration by reducing the exploration space and predicting the application sensitivity using Machine Learning (ML) techniques. A basis for these techniques are implicit correlations between MPI semantics, application context, critical application features, and application responses to faulty collective communications. The experimental results show that our approach reduces the fault injection points and tests by 97% for representative benchmarks (NAS Parallel Benchmarks (NPB)) and a realistic application (Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS)) on a production supercomputer. Further, we statistically generalize the application sensitivity to faulty collective communications for these workloads, and present correlation between application features and the sensitivity.

Computer Science↗
DOE OSTI2020

Distributed non-negative matrix factorization with determination of the number of latent features

The holistic analysis and understanding of the latent (that is, not directly observable) variables and patterns buried in large datasets is crucial for data-driven science, decision making and emergency response. Such exploratory analyses require devising unsupervised learning methods for data mining and extraction of the latent features, and non-negative matrix factorization (NMF) is one of the prominent such methods. NMF is based on compute-intense non-convex constrained minimization, which, for large datasets requires fast and distributed algorithms. However, current parallel implementations of NMF fail to estimate the number of latent features. In practice, identifying these features is both difficult and significant for pattern recognition and latent feature analysis, especially for large dense matrices. Here, we introduce a distributed NMF algorithm coupled with distributed custom clustering followed by a stability analysis on dense data, which we call DnMFk, to determine the number of latent variables. The results on synthetic data and the classical Swimmer data set demonstrate the accuracy of model determination while scaling nearly linearly across multiple processors for large data. Further, we employ DnMFk to determine the number of hidden features from a terabyte matrix.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Machine Learning-Driven Reliability Estimation of PV Inverters Considering Alert-Ambient Variability

Weather-induced spatio-temporal degradation limits outdoor PV inverter lifetime and reliability, necessitating advanced data analysis. This study employs a top-down, data-driven approach utilizing multiple machine learning (ML) algorithms to estimate inverter reliability in a 1.4 MW PV power plant, considering factors such as irradiance, humidity, temperature, time of day, and weather conditions. An extensive alert dataset from 17 identical inverters, including alert types, propagation, and frequency, reveals significant correlations with environmental factors and inverter output power, enabling the construction of a performance reliability model. Dual-stage supervised-ML models are evaluated for accuracy, with the ‘classification-regression’ model by an artificial neural network (ANN) tested on the averaged “Alert-Ambient” dataset, which is outperformed by ‘clustering-regression’ models using random forest (RF) and K-Nearest Neighbors (KNN) on individual inverter datasets. K-means clustering applies principal component analysis to reduce dimensions, achieving improved accuracy beyond the 80% achieved by ANN on the averaged dataset. Second-stage regression estimates inverter reliability with a mean square error of 0.0195 on the averaged dataset and as low as 0.002 on individual inverter datasets using RF. Furthermore, these findings highlight the method's suitability for estimating PV inverter output reliability under ambient conditions, essential for digital twin development and related applications.

14 SOLAR ENERGY↗
DOE OSTI2020

Searching for solitons in heavy-ion reactions near the fermi energy

Solitons are special shape-conserving hydrodynamical solutions that appear in many areas of physics. Here, we explore the existence of such solutions in microscopic descriptions of the heavy ion reaction 12 C + 28 Si → 12 C*+ 7α in the range E/A = 10 – 65 MeV/u. After recognizing the centrality of the collision and time-reversibility as fundamental requirements for the presence of solitonic 12 C, we utilize the Hybrid α-Cluster model for our analysis with a novel methodology. Our results suggest soliton production for E/A = 25 – 45 MeV/u in the forward direction, with a total cross section being at least in the order of a few μb. This, apart from being encouraging for possible experimental studies, maybe connected to possible toroid states of 28 Si at low angular momenta.

Heavy ion reactions↗
DOE OSTI2019

Reconstructing small-scale lenses from the cosmic microwave background temperature fluctuations

Cosmic microwave background (CMB) lensing is a powerful probe of the matter distribution in the Universe. The standard quadratic estimator, which is typically used to measure the lensing signal, is known to be suboptimal for low-noise polarization data from next-generation experiments. In this paper, we explain why the quadratic estimator will also be suboptimal for measuring lensing on very small scales, even for measurements in temperature where this estimator typically performs well. Though maximum likelihood methods could be implemented to improve performance, we explore a much simpler solution, revisiting a previously proposed method to measure lensing that involves a direct inversion of the background gradient. An important application of this simple formalism is the measurement of cluster masses with CMB lensing. We find that directly applying a gradient inversion matched filter to simulated lensed images of the CMB can tighten constraints on cluster masses compared to the quadratic estimator. While the difference is not relevant for existing surveys, for future surveys it can translate to significant improvements in mass calibration for distant clusters, where galaxy lensing calibration is ineffective due to the lack of enough resolved background galaxies. Improvements can be as large as ~50 per cent for a cluster at z = 2 and a next-generation CMB experiment with 1 μK arcmin noise, and over an order of magnitude for lower noise levels. For future surveys, this simple matched filter or gradient inversion method approaches the performance of maximum likelihood methods, at a fraction of the computational cost.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2021

Multisublattice cluster expansion study of short-range ordering in iron-substituted strontium titanate

Owing to the challenges in obtaining realistic atomic configurations in large chemical phase spaces, it is not straightforward to describe structure–property relations in materials exhibiting configurational disorder. One example is iron-substituted strontium titanate (SrTi 1–x Fe x O 3–d , STF), a promising perovskite-derivative cathode material in solid oxide fuel cells that exhibits full solid solubility 0 ≤ x ≤ 1 and a tendency to exhibit short-range order. Here we demonstrate a multisublattice cluster expansion (CE) framework and apply it to STF across the full composition range. The CE approach is distinct from more traditional CE formulations in that clusters are defined explicitly by the chemical species distributed among multiple sublattices, rather than via cluster functions of occupation variables with decoration. The modified CE approach makes it easy to distinguish meaningful chemical interactions that are harder to extract from conventional CE, since for the latter chemical identity in a cluster is expressed as a product of site occupations. The least absolute shrinkage and selection operator (LASSO) is implemented as a regression analysis tool to select key clusters and avoid overfitting. We demonstrate this formulation on STF, and show that it can accurately predict configurational energies in comparison to conventional CE. From the key clusters, we identify that short-range ordering between substitutional Fe and oxygen vacancies (V O ) results in the formation of Fe–VO strings. In addition, we consider the stability of STF through CE-based Monte Carlo (MC) simulations and confirm the presence of superstructures that were previously observed in transmission electron microscopy. In this work, analysis of atomic configurations from MC samples reveals variations in the oxidation state of Fe atoms, which can be explained by the ordering tendency of Fe and V O . The cluster description and selection formalism described here may be applied to other disordered multisublattice systems for accurate and efficient material modeling.

36 MATERIALS SCIENCE↗
NASA NTRS2002

Fluid Dynamics of Bubbly Liquids

Experiments have been performed to study the average flow properties of inertially dominated bubbly liquids which may be described by a novel analysis. Bubbles with high Reynolds number and low Weber number may produce a fluid velocity disturbance that can be approximated by a potential flow. We studied the behavior of suspensions of bubbles of about 1.5 mm diameter in vertical and inclined channels. The suspension was produced using a bank of 900 glass capillaries with inner diameter of about 100 microns in a quasi-steady fashion. In addition, salt was added to the suspension to prevent bubble-bubble coalescence. As a result, a nearly monodisperse suspension of bubble was produced. By increasing the inclination angle, we were able to explore an increasing amount of shear to buoyancy motion. A pipe flow experiment with the liquid being recirculated is under construction. This will provide an even larger range of shear to buoyancy motion. We are planning a microgravity experiment in which a bubble suspension is subjected to shearing in a couette cell in the absence of a buoyancy-driven relative motion of the two phases. By employing a single-wire, hot film anemometer, we were able to obtain the liquid velocity fluctuations. The shear stress at the wall was measured using a hot film probe flush mounted on the wall. The gas volume fraction, bubble velocity, and bubble velocity fluctuations were measured using a homemade, dual impedance probe. In addition, we also employed a high-speed camera to obtain the bubble size distribution and bubble shape in a dilute suspension. A rapid decrease in bubble velocity for a dilute bubble suspension is attributed to the effects of bubble-wall collisions. The more gradual decrease of bubble velocity as gas volume fraction increases, due to subsequent hindering of bubble motion, is in qualitative agreement with the predictions of Spelt and Sangani for the effects of potential-flow bubble-bubble interactions on the mean velocity. The ratio of the bubble velocity variance to the square of the mean is 0(0.1). For these conditions Spelt and Sangani predicted that the homogeneous suspension would be unstable and clustering into horizontal rafts will take place. Evidence for bubble clustering is obtained by analysis of video images. The liquid velocity variance is larger than would be expected for a homogeneous suspension and the liquid velocity frequency spectrum indicates the presence of velocity fluctuations that are slow compared with the time for the passage of an individual bubble. These observations provide further evidence for bubble clustering.

Tsang, Y. H.↗