Publications

We report the energetics and properties of the β, α, R, γ, λ, and δ polymorphs of MnO[subscript 2] within density functional theory, comparing the performance of the recently introduced SCAN functional with that of conventional exchange-correlation functionals and experiment. We find that SCAN uniquely yields accurate formation energies and properties across all MnO[subscript 2] polymorphs. We explain the superior performance of SCAN based on its satisfaction of all known constraints appropriate to a semilocal exchange-correlation functional and its accurate representation of all types of orbital overlap.

The advantage of 3D printing—that is, additive manufacturing (AM) of structural materials—has been severely compromised by their disappointing fatigue properties1,2. Commonly, poor fatigue properties appear to result from the presence of microvoids induced by current printing process procedures3,4. Accordingly, the question that we pose is whether the elimination of such microvoids can provide a feasible solution for marked enhancement of the fatigue resistance of void-free AM (Net-AM) alloys. Here we successfully rebuild an approximate void-free AM microstructure in Ti-6Al-4V titanium alloy by development of a Net-AM processing technique through an understanding of the asynchronism of phase transformation and grain growth. We identify the fatigue resistance of such AM microstructures and show that they lead to a high fatigue limit of around 1 GPa, exceeding the fatigue resistance of all AM and forged titanium alloys as well as that of other metallic materials. We confirm the high fatigue resistance of Net-AM microstructures and the potential advantages of AM processing in the production of structural components with maximum fatigue strength, which is beneficial for further application of AM technologies in engineering fields.

ABSTRACT Supernova (SN) rates serve as an important probe of star formation models and initial mass functions. Near-infrared seeing-limited ground-based surveys typically discover a factor of 3–10 fewer SNe than predicted from far-infrared luminosities owing to sensitivity limitations arising from both a variable point-spread function (PSF) and high dust extinction in the nuclear regions of star-forming galaxies. This inconsistency has potential implications for our understanding of star-formation rates and massive-star evolution, particularly at higher redshifts, where star-forming galaxies are more common. To resolve this inconsistency, a successful SN survey in the local universe must be conducted at longer wavelengths and with a space-based telescope, which has a stable PSF to reduce the necessity for any subtraction algorithms and thus residuals. Here, we report on a 2-yr Spitzer/IRAC 3.6 $\mu$m survey for dust-extinguished SNe in the nuclear regions of forty luminous infrared galaxies (LIRGs) within 200 Mpc. The asymmetric Spitzer PSF results in worse than expected subtraction residuals when implementing standard template subtraction. Forward-modelling techniques improve our sensitivity by several ∼1.5 mag. We report the detection of 9 SNe, five of which were not discovered by optical surveys. After adjusting our predicted rates to account for the sensitivity of our survey, we find that the number of detections is consistent with the models. While this search is none the less hampered by a difficult-to-model PSF and the relatively poor resolution of Spitzer, it will benefit from future missions, such as Roman and the James Webb Space Telescope, with higher resolution and more symmetric PSFs.

The degradation of poly(methyl methacrylate) films exposed to CF4 and CF4/O2 plasmas in a parallel plate reactor is investigated by means of gel permeation chromatography. Profile results indicate that degradation is high in the top few thousand angstroms, although reductions in molecular weight are observed throughout the 1.5 μm films. Oxygen additions to CF4 generate increases in PMMA etch rates, but molecular weight profiles suggest that degradation is still confined to the film surface. The shifts in molecular weight distribution obtained from plasma degradation experiments are simulated by a random scissioning model.

ABSTRACT The origins of quasi-periodic eruptions (QPEs) are poorly understood, although most theoretical explanations invoke an accretion disc around a supermassive black hole. The gas and stellar environments in the galactic nuclei of these sources are also poorly constrained. In this paper, we present an analysis of archival Hubble Space Telescope (HST) images to study the narrow-line [O iii] emission in the QPE source GSN 069. We find strong evidence for a compact nuclear [O iii] emission region of size ≲ 35 pc, overlaid on top of extended [O iii] emission up to 1 kpc away from the nucleus. The age of the accretion system is estimated to be between 10 and 100 yr. The [O iii] luminosity of the compact region was measured to be $(2.1 \pm 0.3) \times 10^{40}\, \rm erg\, s^{-1}$. Based on cloudy simulations, we constrain that the [O iii] emitting gas has a hydrogen number density in the range $2 \times 10^{3} \lt n_{\rm H} \lesssim 10^{8}\, \rm cm^{-3}$ and volume filling factor fV < 4 × 10−3. We suggest that the dense gas in the nuclear region of GSN 069 originates from molecular clouds (with total mass $\gtrsim 3 \times 10^{3}\, {\rm M}_{\odot }$), which are freshly ionized by the soft X-ray photons from the accretion disc. We predict possible evolution of the compact narrow-line region on emission-line diagnostic diagrams, and hence future HST or integral-field unit observations can be used to further pin down the age of this puzzling system.

Many "big data" applications must act on data in real time. Running these applications at ever-larger scales requires parallel platforms that automatically handle faults and stragglers. Unfortunately, current distributed stream processing models provide fault recovery in an expensive manner, requiring hot replication or long recovery times, and do not handle stragglers. We propose a new processing model, discretized streams (D-Streams), that overcomes these challenges. D-Streams enable a parallel recovery mechanism that improves efficiency over traditional replication and backup schemes, and tolerates stragglers. We show that they support a rich set of operators while attaining high per-node throughput similar to single-node systems, linear scaling to 100 nodes, sub-second latency, and sub-second fault recovery. Finally, D-Streams can easily be composed with batch and interactive query models like MapReduce, enabling rich applications that combine these modes. We implement D-Streams in a system called Spark Streaming.

Large Language Model (LLM) serving systems batch concurrent user requests to achieve efficient serving. However, in real-world deployments, such inter-request parallelism from batching is often limited by external factors such as low request rates or memory constraints. Recent works focus on intra-request parallelism from speculative decoding as a solution to this problem. Unfortunately, benefits from intra-request parallelism are often fragile, as speculative decoding causes overhead, and speculated tokens may miss. We observe that speculative decoding may degrade LLM serving performance if added naively without tuning to the incoming requests and the speculation method. To alleviate the need for expert tuning and make speculative decoding more robust, we present TurboSpec, a speculation control system that automatically profiles the execution environment and utilizes a feedback-based algorithm to dynamically adjust the amount of intra-request parallelism in LLM serving. TurboSpec predicts "goodput" - the amount of successfully generated tokens - to evaluate and adjust intra-request parallelism amount to that with the highest goodput in runtime. We implement TurboSpec on a real-world LLM serving system vLLM and demonstrate its effectiveness across diverse workloads and hardware configurations, providing consistent performance improvements across all test scenarios.

Experience froman operational Map-Reduce cluster reveals that outliers significantly prolong job completion. The causes for outliers include run-time contention for processor, memory and other resources, disk failures, varying bandwidth and congestion along network paths and, imbalance in task workload. We present Mantri, a system that monitors tasks and culls outliers using cause- and resource-aware techniques. Mantri's strategies include restarting outliers, network-aware placement of tasks and protecting outputs of valuable tasks. Using real-time progress reports, Mantri detects and acts on outliers early in their lifetime. Early action frees up resources that can be used by subsequent tasks and expedites the job overall. Acting based on the causes and the resource and opportunity cost of actions lets Mantri improve over prior work that only duplicates the laggards. Deployment in Bing's production clusters and trace-driven simulations show that Mantri improves job completion times by 32%.