Lawrence Berkeley National Laboratory

Abstract: The ATLAS Google Project was established as part of an ongoing evaluation of the use of commercial clouds by the ATLAS Collaboration, in anticipation of the potential future adoption of such resources by WLCG grid sites to fulfil or complement their computing pledges. Seamless integration of Google cloud resources into the worldwide ATLAS distributed computing infrastructure was achieved at large scale and for an extended period of time, and hence cloud resources are shown to be an effective mechanism to provide additional, flexible computing capacity to ATLAS. For the first time a total cost of ownership analysis has been performed, to identify the dominant cost drivers and explore effective mechanisms for cost control. Network usage significantly impacts the costs of certain ATLAS workflows, underscoring the importance of implementing such mechanisms. Resource bursting has been successfully demonstrated, whilst exposing the true cost of this type of activity. A follow-up to the project is underway to investigate methods for improving the integration of cloud resources in data-intensive distributed computing environments and reducing costs related to network connectivity, which represents the primary expense when extensively utilising cloud resources.

The discovery of ~80 K superconductivity in nickelate La3Ni2O7 under pressure has ignited intense interest. Here, we present a comprehensive first-principles study of the electron-phonon (e-ph) coupling in La3Ni2O7 and its implications on the observed superconductivity. Our results conclude that the e-ph coupling is too weak (with a coupling constant λ ≲ 0.5) to account for the high Tc, albeit interesting many-electron correlation effects exist. While Coulomb interactions (via GW self-energy and Hubbard U) enhance the e-ph coupling strength, electron doping (oxygen vacancies) introduces no major changes. Additionally, different structural phases display varying characteristics near the Fermi level, but do not alter the conclusion. The e-ph coupling landscape of La3Ni2O7 is intrinsically different from that of infinite-layer nickelates. These findings suggest that a phonon-mediated mechanism is unlikely to be responsible for the observed superconductivity in La3Ni2O7, pointing instead to an unconventional nature.

Growing interest in Artificial Intelligence (AI) has resulted in a surge in demand for faster methods of Machine Learning (ML) model training and inference. This demand for speed has prompted the use of high performance computing (HPC) systems that excel in managing distributed workloads. Because data is the main fuel for AI applications, the performance of the storage and I/O subsystem of HPC systems is critical. In the past, HPC applications accessed large portions of data written by simulations or experiments or ingested data for visualizations or analysis tasks. ML workloads perform small reads spread across a large number of random files. This shift of I/O access patterns poses several challenges to modern parallel storage systems. In this article, we survey I/O in ML applications on HPC systems, and target literature within a 6-year time window from 2019 to 2024. We define the scope of the survey, provide an overview of the common phases of ML, review available profilers and benchmarks, examine the I/O patterns encountered during offline data preparation, training, and inference, and explore I/O optimizations utilized in modern ML frameworks and proposed in recent literature. Lastly, we seek to expose research gaps that could spawn further R&D.

This article proposes an object layout regenerator called Regen which regenerates and removes the object layout dynamically to improve the read performance of applications. Regen first detects frequent access patterns from the I/O requests of the applications. Second, Regen reorganizes the objects and regenerates or preallocates new object layouts according to the identified access patterns. Finally, Regen removes or reuses the obsolete or regenerated object layouts as necessary. As a result, Regen accelerates access to objects by providing a flexible object layout. We implement Regen as a framework on top of Proactive Data Container (PDC) and evaluate it on Cori supercomputer, a production-scale HPC system, by using realistic HPC I/O benchmarks. The experimental results show that Regen improves the I/O performance by up to 16.92× compared with an existing system.

Recent works have implemented machine learning based solutions for many complex classification tasks including pulse shape discrimination in radiation detection. The present work aims to advance the application of machine learning to pulse shape discrimination in neutron detection. A machine learning based neutron-gamma discrimination technique is investigated for various neutron energy distributions produced from DD, DT, (α,n), and spontaneous fission neutron sources. Comprehensive investigations on the training data generation techniques, the impact of the PMT bias, and the discrimination performance are conducted. With the increase of the PMT bias voltage, the neutron classification performance peaked at 1500 V with 81 % of validation neutrons being identified at a false positive rate of 1E-6 while the further bias increase led to a notable degradation in performance. The unsatisfactory classification performance encountered when training off of one neutron source type and classifying neutrons from the other source types was greatly improved with the application of the transfer learning techniques. The remaining variation in the performance was accounted for by the energy dependence of the neutron classification. It was demonstrated that at the 1E-6 FPR specificity level, the events within the region of overlap for neutron and photon populations could be separated, down to a detected energy of 30 keVee. An overall intrinsic neutron detection efficiency of 12.5 % was achieved for the ²⁵²Cf neutron source at a false positive rate of 1E-6.

Artificial Intelligence (AI) applications critically depend on data. Poor-quality data produces inaccurate and ineffective AI models that may lead to incorrect or unsafe use. Evaluation of data readiness is a crucial step in improving the quality and appropriateness of data usage for AI. R&D efforts have been spent on improving data quality. However, standardized metrics for evaluating data readiness for use in AI training are still evolving. In this study, we perform a comprehensive survey of metrics used to verify data readiness for AI training. This survey examines more than 140 papers published by ACM Digital Library, IEEE Xplore, journals such as Nature, Springer, and Science Direct, and online articles published by prominent AI experts. This survey aims to propose a taxonomy of data readiness for AI (DRAI) metrics for structured and unstructured datasets. We anticipate that this taxonomy will lead to new standards for DRAI metrics that would be used for enhancing the quality, accuracy, and fairness of AI training and inference.

The increasing interest in the use of membrane systems to desalinate inland brackish water, agricultural drainage, and industrially produced wastewater demands improved means of predicting desalination system performance under variable feedwater compositions. The interaction among water flow, solute transport, and chemical composition in these systems impacts permeate flux evolution. Here, an established multicomponent reactive transport simulator that accounts for these coupled processes is applied to compute osmotic pressure and permeate fluxes in reverse osmosis (RO) systems. The model is first validated by predicting permeate fluxes for a set of benchtop crossflow experiments subject to a range of feed flow rates and compositions, under fouling and non-fouling conditions. Results compare favorably with measured data that show that solutions with similar total dissolved solids concentrations but different compositions result in different permeate fluxes. The model is then applied to predict permeate fluxes from the desalination of produced waters using a commercial spiral wound RO module. For NaCl-dominant brines, at total dissolved salt concentrations (TDS) below about 70 g/L, permeate fluxes are inversely proportional to water mole fraction as the latter is a reasonable approximation of water activity (i.e. ideal mixing). In the case of Ca–Cl-, Na–CO3- and Na–SO4-dominant brines below about 70 g/L TDS, this relationship does not hold as well and tends to overpredict osmotic pressure and thus underpredict permeate fluxes. However, the opposite becomes true at higher TDS values for typical produced waters. The scaling potential of these waters is also computed by allowing the precipitation of minerals above their saturation limit on the RO membrane. This work demonstrates how reactive transport models developed for the analysis of waters from geological systems can be extended to improve process design, optimization, and control in desalination systems from produced waters and beyond.

Isoprenol (3-methyl-3-buten-1-ol) is a precursor to aviation fuels and other commodity chemicals and can be microbially synthesized from renewable carbon streams. Its production has been demonstrated in Pseudomonas putida KT2440 but its titers, rates, and yields have yet to reach commercially viable levels, potentially due to toxicity to the bacterial chassis. We hypothesized that utilization of Tolerization Adaptive Laboratory Evolution (TALE) would generate P. putida hosts more tolerant to isoprenol and suitable for enhanced production phenotypes. Here, we performed a comprehensive TALE campaign using three strains, the wild-type and two strains lacking subsets of known isoprenol catabolism and transport functions in quadruplicate independently evolved lineages. Several evolved clones from each starting strain displayed robust growth (up to 0.2 h^-1) at 8 g/L of isoprenol, where starting strains could not grow. Whole genome resequencing of the 12 independent strain lineages identified convergent mutations. Reverse engineering each of the four commonly mutated regions individually (gnuR, ttgB-PP_1394, PP_3024-PP_5558, PP_1695) resulted in a partial recovery of the tolerance phenotypes observed in the evolved strains. Additionally, a proteomics-guided deletion of the master motility regulator, fleQ, in an evolved clone alleviated the tolerance vs. production trade-off, restoring isoprenol titers and consumption to levels observed in the starting strains. Collectively, this work demonstrated that an integrated strategy of laboratory evolution and rational engineering was effective to develop robust biofuel production hosts with minimized product toxicity.

While the adsorption properties of transition metal catalysts have been widely studied, leading to the discovery of various scaling relations, descriptors of catalytic activity, and well-established computational models, a similar understanding of semiconductor catalysts has not yet been achieved. In this work, we present a high-throughput density functional theory investigation into the adsorption properties of 5 oxides of interest to the photocatalytic CO2 reduction reaction: TiO2 (rutile and anatase), SrTiO3, NaTaO3, and CeO2. Using a systematic approach, we exhaustively identify unique surfaces and construct adsorption structures to undergo geometry optimizations. We then perform a data-driven analysis, which reveals the presence of weak adsorption energy scaling relations, the propensity of adsorbates of interest to interact with oxygen surface sites, and the importance of slab deformation upon adsorption. Our findings are presented in the context of experimental observations and in comparison to previously studied classes of catalysts, such as pure metals and tellurium-containing semiconductors, and reinforce the need for a comprehensive approach to the study of site-specific surface phenomena on semiconductors.