Search

The recent detection of extended $\gamma$-ray emission around middle-aged pulsars is interpreted as inverse-Compton scattering of ambient photons by electron-positron pairs escaping the pulsar wind nebula, which are confined near the system by unclear mechanisms. This emerging population of $\gamma$-ray sources was first discovered at TeV energies and remains underexplored in the GeV range. To address this, we conducted a systematic search for extended sources along the Galactic plane using 14 years of Fermi-LAT data above 10 GeV, aiming to identify a number of pulsar halo candidates and extend our view to lower energies. The search covered the inner Galactic plane ($\lvert l\rvert\leq$ 100$^{\circ}$, $\lvert b\rvert\leq$ 1$^{\circ}$) and the positions of known TeV sources and bright pulsars, yielding broader astrophysical interest. We found 40 such sources, forming the Second Fermi Galactic Extended Sources Catalog (2FGES), most with 68% containment radii smaller than 1.0$^{\circ}$ and relatively hard spectra with photon indices below 2.5. We assessed detection robustness using field-specific alternative interstellar emission models and by inspecting significance maps. Noting 13 sources previously known as extended in the 4FGL-DR3 catalog and five dubious sources from complex regions, we report 22 newly detected extended sources above 10 GeV. Of these, 13 coincide with H.E.S.S., HAWC, or LHAASO sources; six coincide with bright pulsars (including four also coincident with TeV sources); six are associated with 4FGL point sources only; and one has no association in the scanned catalogs. Notably, six to eight sources may be related to pulsars as classical pulsar wind nebulae or pulsar halos., Comment: 43 pages, 17 figures, 11 tables

We present a comprehensive study of the C-metric in $2+1$ dimensions, placing it within a shell of stress energy and matching it to an exterior vacuum AdS metric. The $2+1$ C-metric is not circularly symmetric and hence neither are the constructed shells, which instead take on a cuspoidal or teardrop shape. We interpret the stress energy of the shells as a perfect fluid, calculating the energy density and pressure. For accelerating particles (Class I), we find the stress energy is concentrated on the part of shell farthest from the direction of acceleration and always respects the strong energy condition. For accelerating black holes (Class I$_{\mathrm{C}}$, II, and III), the shell stress energy may either respect or violate the energy conditions depending on the parameter of the exterior metric -- between the two regimes lies a critical value of the external parameter for which the shell stress energy vanishes, leading to new solutions of Einstein's field equations, which fall into three categories: an accelerated black hole pulled by a finite-length string with a point particle at the other end, an accelerated black hole pushed by a finite-length strut with a point particle at the other end, and an accelerated black hole pushed from one side by a finite-length strut and pulled from the other by a finite-length string, each with a point particle at the other end., Comment: 43 pages, 24 figures; v2 reference added

The Mg ii h&k lines are key diagnostics of the solar chromosphere. They are sensitive to the temperature, density, and non-thermal velocities in the chromosphere. The average Mg ii h&k line profiles arising from previous 3D chromospheric simulations are too narrow. We study the formation and properties of the Mg ii h&k lines in a model atmosphere. We also compare the average spectrum, peak intensity, and peak separation of Mg ii k with a representative observation taken by IRIS. We use a model based on the recently developed non-equilibrium version of the radiative magneto-hydrodynamics code MURaM, in combination with forward modeling using the radiative transfer code RH1.5D to obtain synthetic spectra. Our model resembles an enhanced network region created by using an evolved MURaM quiet sun simulation and adding a similar imposed large-scale bipolar magnetic field as in the public Bifrost snapshot of a bipolar magnetic feature. The line width and the peak separation of the spatially averaged spectrum of the Mg ii h&k lines from the MURaM simulation are close to a representative observation from the quiet sun which also includes network fields. However, we find the synthesized line width to be still slightly narrower than in the observation. We find that velocities in the chromosphere play a dominant role in the broadening of the spectral lines. While the average synthetic spectrum also shows a good match with the observations in the pseudo continuum between the two emission lines, the peak intensities are higher in the modeled spectrum. This discrepancy may partly be due to the larger magnetic flux density in the simulation than in the considered observations but also due to the 1.5D radiative transfer approximation. Our findings show that strong maximum velocity differences or turbulent velocities in the chromosphere are necessary to reproduce the observed line widths., Comment: 17 pages, 13 figures; accepted for publication by A&A

The interaction of an ultra-intense laser pulse with a near critical density target results in the formation of a plasma channel, a strong azimuthal magnetic field and moving vortices. An application of this is the generation of energetic and collimated ion beams via Magnetic Vortex Acceleration. The optimized regime of Magnetic Vortex Acceleration is becoming experimentally accessible with new high intensity laser beamlines coming online and advances made in near critical density target fabrication. The robustness of the acceleration mechanism with realistic experimental conditions is examined with three-dimensional simulations. Of particular interest is the acceleration performance with different laser temporal contrast conditions, in some cases leading to pre-expanded target profiles prior to the arrival of the main pulse. Preplasma effects on the structure of the accelerating fields is explored, including a detailed analysis of the ion beam properties and the efficiency of the process. Improved scaling laws for the MVA mechanism, including the laser focal spot size effects, are presented., Comment: 12 pages, 7 figures

We present a complete analysis of Fermi Large Area Telescope (LAT) data of GRB 221009A, the brightest Gamma-Ray Burst (GRB) ever detected. The burst emission above 30 MeV detected by the LAT preceded by 1 s the low-energy (< 10 MeV) pulse that triggered the Fermi Gamma-Ray Burst Monitor (GBM), as has been observed in other GRBs. The prompt phase of GRB 221009A lasted a few hundred seconds. It was so bright that we identify a Bad Time Interval (BTI) of 64 seconds caused by the extremely high flux of hard X-rays and soft gamma rays, during which the event reconstruction efficiency was poor and the dead time fraction quite high. The late-time emission decayed as a power law, but the extrapolation of the late-time emission during the first 450 seconds suggests that the afterglow started during the prompt emission. We also found that high-energy events observed by the LAT are incompatible with synchrotron origin, and, during the prompt emission, are more likely related to an extra component identified as synchrotron self-Compton (SSC). A remarkable 400 GeV photon, detected by the LAT 33 ks after the GBM trigger and directionally consistent with the location of GRB 221009A, is hard to explain as a product of SSC or TeV electromagnetic cascades, and the process responsible for its origin is uncertain. Because of its proximity and energetic nature, GRB 221009A is an extremely rare event., Comment: 60 pages, 38 figures, 9 tables

Everyday AI detection requires differentiating between people and AI in informal, online conversations. In many cases, people will not interact directly with AI systems but instead read conversations between AI systems and other people. We measured how well people and large language models can discriminate using two modified versions of the Turing test: inverted and displaced. GPT-3.5, GPT-4, and displaced human adjudicators judged whether an agent was human or AI on the basis of a Turing test transcript. We found that both AI and displaced human judges were less accurate than interactive interrogators, with below chance accuracy overall. Moreover, all three judged the best-performing GPT-4 witness to be human more often than human witnesses. This suggests that both humans and current LLMs struggle to distinguish between the two when they are not actively interrogating the person, underscoring an urgent need for more accurate tools to detect AI in conversations.

This article presents constraints on dark-matter-electron interactions obtained from the first underground data-taking campaign with multiple SuperCDMS HVeV detectors operated in the same housing. An exposure of 7.63 g-days is used to set upper limits on the dark-matter-electron scattering cross section for dark matter masses between 0.5 and 1000 MeV/$c^2$, as well as upper limits on dark photon kinetic mixing and axion-like particle axioelectric coupling for masses between 1.2 and 23.3 eV/$c^2$. Compared to an earlier HVeV search, sensitivity was improved as a result of an increased overburden of 225 meters of water equivalent, an anticoincidence event selection, and better pile-up rejection. In the case of dark-matter-electron scattering via a heavy mediator, an improvement by up to a factor of 25 in cross-section sensitivity was achieved., Comment: 7 pages + title and references, 4 figures, and 1 table

Anti-de Sitter spacetime and the static patch of de Sitter spacetime are arenas for investigating thermal and finite-size effects seen by an accelerated quantum observer. We consider an Unruh-DeWitt detector in uniform circular motion coupled to a conformal scalar field in $(2+1)$-dimensional de Sitter and anti-de Sitter spacetimes in the limit of a small cosmological constant $\Lambda$. In anti-de Sitter spacetime, where $\Lambda$ mimics spatial confinement, we find that the resonance peaks in the detector's response closely match those of a detector in Minkowski space with a cylindrical boundary, but with curvature corrections, more significant when the field has an ambient temperature. In the static patch of de Sitter spacetime, in the Euclidean vacuum, we show that the leading curvature correction to the detector's response is proportional to $\Lambda$, as in zero temperature anti-de Sitter, whilst the temperature corrections decay exponentially as $\Lambda \to 0$., Comment: 18 pages: 10 pages + 8 pages appendices. 1 figure

Recent empirical results have sparked a debate about whether or not Large Language Models (LLMs) are capable of Theory of Mind (ToM). While some have found LLMs to be successful on ToM evaluations such as the False Belief task (Kosinski, 2023), others have argued that LLMs solve these tasks by exploiting spurious correlations -- not representing beliefs -- since they fail on trivial alterations to these tasks (Ullman, 2023). In this paper, we introduce SCALPEL: a technique to generate targeted modifications for False Belief tasks to test different specific hypotheses about why LLMs fail. We find that modifications which make explicit common inferences -- such as that looking at a transparent object implies recognizing its contents -- preserve LLMs' performance. This suggests that LLMs' failures on modified ToM tasks could result from a lack of more general commonsense reasoning, rather than a failure to represent mental states. We argue that SCALPEL could be helpful for explaining LLM successes and failures in other cases.

We evaluated 3 systems (ELIZA, GPT-3.5 and GPT-4) in a randomized, controlled, and preregistered Turing test. Human participants had a 5 minute conversation with either a human or an AI, and judged whether or not they thought their interlocutor was human. GPT-4 was judged to be a human 54% of the time, outperforming ELIZA (22%) but lagging behind actual humans (67%). The results provide the first robust empirical demonstration that any artificial system passes an interactive 2-player Turing test. The results have implications for debates around machine intelligence and, more urgently, suggest that deception by current AI systems may go undetected. Analysis of participants' strategies and reasoning suggests that stylistic and socio-emotional factors play a larger role in passing the Turing test than traditional notions of intelligence., Comment: 23 pages, 13 figures

Structural disturbances, such as galaxy mergers or instabilities, are key candidates for driving galaxy evolution, so it is important to detect and quantify galaxies hosting these disturbances spanning a range of masses, environments, and cosmic times. Traditionally, this is done by quantifying the asymmetry of a galaxy as part of the concentration-asymmetry-smoothness system, $A_{\rm{CAS}}$, and selecting galaxies above a certain threshold as merger candidates. However, in this work, we show that $A_{\rm{CAS}}$, is extremely dependent on imaging properties -- both resolution and depth -- and thus defining a single $A_{\rm{CAS}}$ threshold is impossible. We analyze an alternative root-mean-squared asymmetry, $A_{\rm{RMS}}$, and show that it is independent of noise down to the average SNR per pixel of 1. However, both metrics depend on the resolution. We argue that asymmetry is, by design, always a scale-dependent measurement, and it is essential to define an asymmetry at a given physical resolution, where the limit should be defined by the size of the smallest features one wishes to detect. We measure asymmetry of a set of $z\approx0.1$ galaxies observed with HST, HSC, and SDSS, and show that after matching the resolution of all images to 200 pc, we are able to obtain consistent $A_{\rm{RMS, 200pc}}$ measurements with all three instruments despite the vast differences in the original resolution or depth. We recommend that future studies use $A_{\rm{RMS, x pc}}$ measurement when evaluating asymmetry, where $x$ is defined by the physical size of the features of interest, and is kept consistent across the dataset, especially when the redshift or image properties of galaxies in the dataset vary., Comment: 26 pages, 10 figures. Accepted to the Open Journal for Astrophysics

Observed spectral profiles of emission lines from the corona are found to have widths exceeding the thermal line width. To investigate the physical mechanism, we run a 3D MHD model of a single, straightened loop in which we partially resolve turbulent motions that form in response to the driving by self-consistently evolving magneto-convection in the photosphere. The convective motions shear and twist the magnetic field lines, leading to heating. From the model we synthesize spectral profiles of emission lines forming at temperatures around and above 1 MK. The coronal heating process generates a range of velocity amplitudes and directions structured on a scale much smaller than the resolving power of current instruments, leading to a broadening of the spectral lines. Our model includes the mass exchange between corona and chromosphere, thus we also capture flows parallel to the loop axis. We find that the spectral lines show a non-thermal line broadening roughly consistent with observations for a viewing angle perpendicular to the axis. The broadening through field-parallel flows is comparable, although slightly smaller. The line broadening is independent of the instrument resolution for a perpendicular line-of-sight. We can connect the non-thermal line broadening to heating events and flows. While small-scale velocities along the line-of-sight are mainly responsible for the broadening observed perpendicular to the loop, chromospheric evaporation is important for the line broadening observed along the loop. The model reproduces observed values for non-thermal line widths. In the model these result from continuous driving by magnetoconvection, without imposing driving motions or starting from an already braided field., Comment: 18 pages, 18 figures

Low precision training can significantly reduce the computational overhead of training deep neural networks (DNNs). Though many such techniques exist, cyclic precision training (CPT), which dynamically adjusts precision throughout training according to a cyclic schedule, achieves particularly impressive improvements in training efficiency, while actually improving DNN performance. Existing CPT implementations take common learning rate schedules (e.g., cyclical cosine schedules) and use them for low precision training without adequate comparisons to alternative scheduling options. We define a diverse suite of CPT schedules and analyze their performance across a variety of DNN training regimes, some of which are unexplored in the low precision training literature (e.g., node classification with graph neural networks). From these experiments, we discover alternative CPT schedules that offer further improvements in training efficiency and model performance, as well as derive a set of best practices for choosing CPT schedules. Going further, we find that a correlation exists between model performance and training cost, and that changing the underlying CPT schedule can control the tradeoff between these two variables. To explain the direct correlation between model performance and training cost, we draw a connection between quantized training and critical learning periods, suggesting that aggressive quantization is a form of learning impairment that can permanently damage model performance., Comment: 20 pages, 8 figures, 1 table, ACML 2023

We evaluated GPT-4 in a public online Turing test. The best-performing GPT-4 prompt passed in 49.7% of games, outperforming ELIZA (22%) and GPT-3.5 (20%), but falling short of the baseline set by human participants (66%). Participants' decisions were based mainly on linguistic style (35%) and socioemotional traits (27%), supporting the idea that intelligence, narrowly conceived, is not sufficient to pass the Turing test. Participant knowledge about LLMs and number of games played positively correlated with accuracy in detecting AI, suggesting learning and practice as possible strategies to mitigate deception. Despite known limitations as a test of intelligence, we argue that the Turing test continues to be relevant as an assessment of naturalistic communication and deception. AI models with the ability to masquerade as humans could have widespread societal consequences, and we analyse the effectiveness of different strategies and criteria for judging humanlikeness., Comment: 28 pages, 21 figures

It is well known that the experience of a linearly accelerated observer with acceleration $a$, interacting with a massless scalar field in its vacuum state in $3+1$ Minkowski spacetime, is identical to that of a static observer interacting with a massless scalar field in a thermal state of temperature $a/2\pi$ in $3+1$ Minkowski spacetime. We study the robustness of this duality by comparing an observer undergoing circular motion in a thermal bath with an observer that undergoes circular motion around a linearly accelerated trajectory. We find that in most regimes, observers in these two cases experience the field in different ways, and are generally able to tell the difference between the two cases by measuring observables localized along their trajectories., Comment: 19 pages, 3 figures. v2: sections 2 and 6 expanded after referee comments. v3: typos corrected. v4: added post-publication note

We determine the conjugacy classes of the Poincar\'e group $\mathrm{ISO}^+(n,1)$ and apply this to classify the stationary trajectories of Minkowski spacetimes in terms of timelike Killing vectors. Stationary trajectories are the orbits of timelike Killing vectors and, equivalently, the solutions to Frenet-Serret equations with constant curvature coefficients. We extend the $3+1$ Minkowski spacetime Frenet-Serret equations due to Letaw to Minkowski spacetimes of arbitrary dimension. We present the explicit families of stationary trajectories in $4+1$ Minkowski spacetime., Comment: 13 pages. v2: Added sentence to Section II B 2. after referee comments

We investigate whether the Babcock-Leighton flux-transport dynamo model remains in agreement with observations if the meridional flow profile is taken from helioseismic inversions. Additionally, we investigate the effect of the loss of toroidal flux through the solar surface. We employ the 2D flux-transport BL dynamo framework. We use the helioseismically-inferred meridional flow profile, and include toroidal flux loss in a way that is consistent with the amount of poloidal flux generated by Joy's law. Our model does not impose a preference for emergences at low latitudes, but we do require that the model produces such a preference. We can find solutions in general agreement with observations, including the equatorward drift of the butterfly wings and the cycle's 11 year period. The most important free parameters in the model are the depth to which the radial turbulent pumping extends and the turbulent diffusivity in the lower half of the convection zone. We find that the pumping needs to extend to depths of about $0.80R_{\odot}$ and the bulk turbulent diffusivity needs to be around 10 km$^2$/s or less. We find that the emergences are restricted to low latitudes without the need to impose such a preference. The flux-transport BL model, incorporating the helioseismically inferred meridional flow and toroidal field loss term, is compatible with the properties of the observed butterfly diagram and with the observed toroidal loss rate. Reasonably tight constraints are placed on the remaining free parameters. The pumping needs to be to just below the depth corresponding to the location where the meridional flow changes direction. Our linear model does not however reproduce the observed "rush to the poles" of the diffuse surface radial field resulting from the decay of sunspots -- reproducing this might require the imposition of a preference for flux to emerge near the equator., Comment: 10 pages, 8 figures, accepted by A&A

Vortex flows have been found in the photosphere, chromosphere and low corona in observations and simulations. It has been suggested that vortices play an important role for channeling energy and plasma into the corona, but the impact of vortex flows on the corona has not directly been studied in a realistic setup. We investigate the role vortices play for coronal heating using high resolution simulations of coronal loops. The vortices are not artificially driven, but arise self-consistently from magnetoconvection. We perform 3D resistive MHD simulations with the MURaM code. Studying an isolated coronal loop in a Cartesian geometry allows us to resolve the structure of the loop interior. We conduct a statistical analysis to determine vortex properties as a function of height from the chromosphere into the corona. We find that the energy injected into the loop is generated by internal coherent motions within strong magnetic elements. A significant part of the resulting Poynting flux is channeled through the chromosphere in vortex tubes forming a magnetic connection between the photosphere and corona. Vortices can form contiguous structures that reach up to coronal heights, but in the corona itself the vortex tubes get deformed and eventually lose their identity with increasing height. Vortices show increased upward directed Poynting flux and heating rate both in the chromosphere and corona, but their effect becomes less pronounced with increasing height. While vortices play an important role for the energy transport and structuring in the chromosphere and low corona, their importance higher up in the atmosphere is less clear since the swirls are less distinguishable from their environment. Vortex tubes reaching the corona show a complex relationship with the coronal emission., Comment: 18 pages, 12 figures

Analogue gravity offers an approach for testing the universality and robustness of quantum field theories in curved spacetimes and validating them using down-to-earth, laboratory-based experiments. Fluid interfaces are a promising framework for creating these gravity simulators and have successfully replicated phenomena such as Hawking radiation and black hole superradiance. Recent work has shown that hydrodynamical instabilities on the interface between two fluids can capture features of the post-inflationary thermalisation of the Early Universe. In this study, we extend fluid dynamics methods to develop an effective field theory for the interface between two fluids, demonstrating the equivalence between the governing equations and a relativistic scalar field in an analogue spacetime. We also show that the interfacial height field serves as the analogue relativistic field even in a nonlinear, interacting field theory. We propose that these mathematical equivalences can be extrapolated to probe regimes where calculations are challenging or impractical. Our work provides a new framework for simulating far-from-equilibrium cosmological and gravitational scenarios in the laboratory., Comment: 17 pages, contribution to the proceedings of the "Avenues of Quantum Field Theory in Curved Space-time" conference, Genova, September 2022

We report the discovery of GRB 221009A, the highest flux gamma-ray burst ever observed by the Fermi Gamma-ray Burst Monitor (GBM). This GRB has continuous prompt emission lasting more than 600 seconds which smoothly transitions to afterglow visible in the GBM energy range (8 keV--40 MeV), and total energetics higher than any other burst in the GBM sample. By using a variety of new and existing analysis techniques we probe the spectral and temporal evolution of GRB 221009A. We find no emission prior to the GBM trigger time (t0; 2022 October 9 at 13:16:59.99 UTC), indicating that this is the time of prompt emission onset. The triggering pulse exhibits distinct spectral and temporal properties suggestive of the thermal, photospheric emission of shock-breakout, with significant emission up to $\sim$15 MeV. We characterize the onset of external shock at t0+600 s and find evidence of a plateau region in the early-afterglow phase which transitions to a slope consistent with Swift-XRT afterglow measurements. We place the total energetics of GRB 221009A in context with the rest of the GBM sample and find that this GRB has the highest total isotropic-equivalent energy ($\textrm{E}_{\gamma,\textrm{iso}}=1.0\times10^{55}$ erg) and second highest isotropic-equivalent luminosity ($\textrm{L}_{\gamma,\textrm{iso}}=9.9\times10^{53}$ erg/s) based on redshift of z = 0.151. These extreme energetics are what allowed us to observe the continuously emitting central engine of GBM from the beginning of the prompt emission phase through the onset of early afterglow., Comment: 26 pages 7 figures - accepted for publication in ApJL

The Unruh effect states that a uniformly linearly accelerated observer with proper acceleration $a$ experiences the Minkowski vacuum as a thermal state at temperature $T_U=a/(2\pi)$. An observer in uniform circular motion experiences a similar effective temperature, operationally defined in terms of excitation and de-excitation rates, and physically interpretable in terms of synchrotron radiation, but this effective temperature depends not just on the acceleration but also on the orbital speed and the excitation energy. In this paper we consider an observer in uniform circular motion when the Minkowski vacuum is replaced by an ambient thermal bath, and we address the interplay of ambient temperature, Doppler effect, acceleration, and excitation energy. Specifically, we consider a massless scalar field in $2 + 1$ spacetime dimensions, probed by an Unruh-DeWitt detector, in a Minkowski (rather than proper) time formulation: this setting describes proposed analogue spacetime systems in which the effect may become experimentally testable, and in which an ambient temperature will necessarily be present. We establish analytic results for the observer's effective temperature in several asymptotic regions of the parameter space and provide numerical results in the interpolating regions, finding that an acceleration effect can be identified even when the Doppler effect dominates the overall magnitude of the response. We also identify parameter regimes where the observer sees a temperature lower than the ambient temperature, experiencing a cooling Unruh effect., Comment: 24 pages, 5 figures. v2: references added, typos corrected. v3: references added, section 2 expanded after referee comments, typos corrected

We measured the nuclear--recoil ionization yield in silicon with a cryogenic phonon-sensitive gram-scale detector. Neutrons from a mono-energetic beam scatter off of the silicon nuclei at angles corresponding to energy depositions from 4\,keV down to 100\,eV, the lowest energy probed so far. The results show no sign of an ionization production threshold above 100\,eV. These results call for further investigation of the ionization yield theory and a comprehensive determination of the detector response function at energies below the keV scale.

An accelerated observer moving through empty space sees particles appearing and disappearing, while an observer with a constant velocity does not register any particles. This phenomenon, generally known as the Unruh effect, relies on an initial vacuum state, thereby unifying the experience of all inertial observers. We propose an experiment to probe this observer-dependent detector response, using a laser beam in circular motion as a local detector of superfluid helium-4 surface modes or third sound waves. To assess experimental feasibility, we develop a theoretical framework to include a non-zero temperature initial state. We find that an acceleration-dependent signal persists, independent of the initial temperature. By introducing a signal-to-noise measure we show that observing this signal is within experimental reach., Comment: 18 pages, 3 figures. v2: Figure 2 axis amended, typos corrected. v3: Added section 4 after referee comments, references added, author order updated, typos corrected

We present a new analysis of previously published of SuperCDMS data using a profile likelihood framework to search for sub-GeV dark matter (DM) particles through two inelastic scattering channels: bremsstrahlung radiation and the Migdal effect. By considering these possible inelastic scattering channels, experimental sensitivity can be extended to DM masses that are undetectable through the DM-nucleon elastic scattering channel, given the energy threshold of current experiments. We exclude DM masses down to $220~\textrm{MeV}/c^2$ at $2.7 \times 10^{-30}~\textrm{cm}^2$ via the bremsstrahlung channel. The Migdal channel search provides overall considerably more stringent limits and excludes DM masses down to $30~\textrm{MeV}/c^2$ at $5.0 \times 10^{-30}~\textrm{cm}^2$., Comment: Submitted to PRD

The length of the solar activity cycle fluctuates considerably. The temporal evolution of the corresponding cycle phase, that is, the deviation of the epochs of activity minima or maxima from strict periodicity, provides relevant information concerning the physical mechanism underlying the cyclic magnetic activity. An underlying strictly periodic process (akin to a perfect "clock"), with the observer seeing a superposition of the perfect clock and a small random phase perturbation, leads to long-term phase stability in the observations. Such behavior would be expected if cycles were synchronized by tides caused by orbiting planets or by a hypothetical torsional oscillation in the solar radiative interior. Alternatively, in the absence of such synchronization, phase fluctuations accumulate and a random walk of the phase ensues, which is a typical property of randomly perturbed dynamo models. Based on the sunspot record and the reconstruction of solar cycles from cosmogenic C14, we carried out rigorous statistical tests in order to decipher whether there exists phase synchronization or random walk. Synchronization is rejected at significance levels of between 95% (28 cycles from sunspot data) and beyond 99% (84 cycles reconstructed from C14, while the existence of random walk in the phases is consistent with all data sets. This result strongly supports randomly perturbed dynamo models with little inter-cycle memory., Comment: Astronomy & Astrophysics, in press

Muscle tissue drives nearly all movement in the animal kingdom, providing power, mobility, and dexterity. Technologies for measuring muscle tissue motion, such as sonomicrometry, fluoromicrometry, and ultrasound, have significantly advanced our understanding of biomechanics. Yet, the field lacks the ability to monitor muscle tissue motion for animal behavior outside the lab. Towards addressing this issue, we previously introduced magnetomicrometry, a method that uses magnetic beads to wirelessly monitor muscle tissue length changes, and we validated magnetomicrometry via tightly-controlled in situ testing. In this study we validate the accuracy of magnetomicrometry against fluoromicrometry during untethered running in an in vivo turkey model. We demonstrate real-time muscle tissue length tracking of the freely-moving turkeys executing various motor activities, including ramp ascent and descent, vertical ascent and descent, and free roaming movement. Given the demonstrated capacity of magnetomicrometry to track muscle movement in untethered animals, we feel that this technique will enable new scientific explorations and an improved understanding of muscle function. -- -- El tejido muscular es el motor de casi todos los movimientos del reino animal, ya que proporciona fuerza, movilidad y destreza. Las tecnolog\'ias para medir el movimiento del tejido muscular, como la sonomicrometr\'ia, la fluoromicrometr\'ia y el ultrasonido, han avanzado considerablemente la comprensi\'on de la biomec\'anica. Sin embargo, este campo carece de la capacidad de rastrear el movimiento del tejido muscular en el comportamiento animal fuera del laboratorio. Para abordar este problema, presentamos previamente la magnetomicrometr\'ia, un m\'etodo que utiliza peque\~nos imanes para rastrear de forma inal\'ambrica los cambios de longitud del tejido muscular, y validamos la magnetomicrometr\'ia mediante pruebas estrechamente controladas in situ. En este estudio validamos la precisi\'on de la magnetomicrometr\'ia en comparaci\'on con la fluoromicrometr\'ia usando un modelo de pavo in vivo mientras corre libremente. Demostramos el rastreo en tiempo real de la longitud del tejido muscular de los pavos que se mueven libremente ejecutando varias actividades motoras, incluyendo el ascenso y el descenso en rampa, el ascenso y el descenso vertical, y el movimiento libre. Dada la capacidad demostrada de la magnetomicrometr\'ia para rastrear el movimiento muscular en animales en un contexto m\'ovil, creemos que esta t\'ecnica permitir\'a nuevas exploraciones cient\'ificas y una mejor comprensi\'on de la funci\'on muscular., Comment: in Spanish language. Translation of the postprint, with the published version in English appended to the end of the PDF. Shared First Authorship: Cameron R. Taylor and Seong Ho Yeon; Shared Senior and Corresponding Authorship: Thomas J. Roberts and Hugh M. Herr

The ability to dynamically adapt neural networks to newly-available data without performance deterioration would revolutionize deep learning applications. Streaming learning (i.e., learning from one data example at a time) has the potential to enable such real-time adaptation, but current approaches i) freeze a majority of network parameters during streaming and ii) are dependent upon offline, base initialization procedures over large subsets of data, which damages performance and limits applicability. To mitigate these shortcomings, we propose Cold Start Streaming Learning (CSSL), a simple, end-to-end approach for streaming learning with deep networks that uses a combination of replay and data augmentation to avoid catastrophic forgetting. Because CSSL updates all model parameters during streaming, the algorithm is capable of beginning streaming from a random initialization, making base initialization optional. Going further, the algorithm's simplicity allows theoretical convergence guarantees to be derived using analysis of the Neural Tangent Random Feature (NTRF). In experiments, we find that CSSL outperforms existing baselines for streaming learning in experiments on CIFAR100, ImageNet, and Core50 datasets. Additionally, we propose a novel multi-task streaming learning setting and show that CSSL performs favorably in this domain. Put simply, CSSL performs well and demonstrates that the complicated, multi-step training pipelines adopted by most streaming methodologies can be replaced with a simple, end-to-end learning approach without sacrificing performance., Comment: 52 pages, 7 figures, pre-print

All cool main sequence stars including our Sun are thought to have magnetic fields. Observations of the Sun revealed that even in quiet regions small-scale turbulent magnetic fields are present. Simulations further showed that such magnetic fields affect the subsurface and photospheric structure, and thus the radiative transfer and emergent flux. Since small-scale turbulent magnetic fields on other stars cannot be directly observed, it is imperative to study their effects on the near surface layers numerically. Until recently comprehensive three-dimensional simulations capturing the effect of small-scale turbulent magnetic fields only exists for the solar case. A series of investigations extending SSD simulations for other stars has been started. Here we aim to examine small-scale turbulent magnetic fields in stars of solar effective temperature but different metallicity. We investigate the properties of three-dimensional simulations of the magneto-convection in boxes covering the upper convection zone and photosphere carried out with the MURaM code for metallicity values of $ \rm M/H = \{-1.0, 0.0, 0.5\}$ with and without a small-scale-dynamo. We find that small-scale turbulent magnetic fields enhanced by a small-scale turbulent dynamo noticeably affect the subsurface dynamics and significantly change the flow velocities in the photosphere. Moreover, significantly stronger magnetic field strengths are present in the convection zone for low metallicity. Whereas, at the optical surface the averaged vertical magnetic field ranges from 64G for M/H = 0.5 to 85G for M/H = -1.0., Comment: 13 pages, 18 figures, submitted to A&A

Surfaces of the Sun and other cool stars are filled with magnetic fields, which are either seen as dark compact spots or more diffuse bright structures like faculae. Both hamper detection and characterisation of exoplanets, affecting stellar brightness and spectra, as well as transmission spectra. However, the expected facular and spot signals in stellar data are quite different, for instance they have distinct temporal and spectral profiles. Consequently, corrections of stellar data for magnetic activity can greatly benefit from the insight on whether the stellar signal is dominated by spots or faculae. Here, we utilise a surface flux transport model (SFTM) to show that more effective cancellation of diffuse magnetic flux associated with faculae leads to spot area coverages increasing faster with stellar magnetic activity than that by faculae. Our calculations explain the observed dependence between solar spot and facular area coverages and allow its extension to stars more active than the Sun. This extension enables anticipating the properties of stellar signal and its more reliable mitigation, leading to a more accurate characterisation of exoplanets and their atmospheres., Comment: 10 pages, 6 Figures, accepted for publication in ApJL

The twist of the magnetic field above a sunspot is an important quantity in solar physics. For example, magnetic twist plays a role in the initiation of flares and coronal mass ejections (CMEs). Various proxies for the twist above the photosphere have been found using models of uniformly twisted flux tubes, and are routinely computed from single photospheric vector magnetograms. One class of proxies is based on $\alpha_z$, the ratio of the vertical current to the vertical magnetic field. Another class of proxies is based on the so-called twist density, $q$, which depends on the ratio of the azimuthal field to the vertical field. However, the sensitivity of these proxies to temporal fluctuations of the magnetic field has not yet been well characterized. We aim to determine the sensitivity of twist proxies to temporal fluctuations in the magnetic field as estimated from time-series of SDO/HMI vector magnetic field maps. To this end, we introduce a model of a sunspot with a peak vertical field of 2370 Gauss at the photosphere and a uniform twist density $q= -0.024$ Mm$^{-1}$. We add realizations of the temporal fluctuations of the magnetic field that are consistent with SDO/HMI observations, including the spatial correlations. Using a Monte-Carlo approach, we determine the robustness of the different proxies to the temporal fluctuations. The temporal fluctuations of the three components of the magnetic field are correlated for spatial separations up to 1.4 Mm (more than expected from the point spread function alone). The Monte-Carlo approach enables us to demonstrate that several proxies for the twist of the magnetic field are not biased in each of the individual magnetograms. The associated random errors on the proxies have standard deviations in the range between $0.002$ and $0.006$ Mm$^{-1}$, which is smaller by approximately one order of magnitude than the mean value of $q$., Comment: accepted in A&A

CDMSlite Run 2 was a search for weakly interacting massive particles (WIMPs) with a cryogenic 600 g Ge detector operated in a high-voltage mode to optimize sensitivity to WIMPs of relatively low mass from 2 - 20 GeV/$c^2$. In this article, we present an effective field theory (EFT) analysis of the CDMSlite Run 2 data using an extended energy range and a comprehensive treatment of the expected background. A binned likelihood Bayesian analysis was performed on the recoil energy data, taking into account the parameters of the EFT interactions and optimizing the data selection with respect to the dominant background components. Energy regions within 5$\sigma$ of known activation peaks were removed from the analysis. The Bayesian evidences resulting from the different operator hypotheses show that the CDMSlite Run 2 data are consistent with the background-only models and do not allow for a signal interpretation assuming any additional EFT interaction. Consequently, upper limits on the WIMP mass and coupling-coefficient amplitudes and phases are presented for each EFT operator. These limits improve previous CDMSlite Run 2 bounds for WIMP masses above 5 GeV/$c^2$., Comment: 16 pages, 8 figures

Cosmic rays are mostly composed of protons accelerated to relativistic speeds. When those protons encounter interstellar material, they produce neutral pions which in turn decay into gamma rays. This offers a compelling way to identify the acceleration sites of protons. A characteristic hadronic spectrum, with a low-energy break around 200 MeV, was detected in the gamma-ray spectra of four Supernova Remnants (SNRs), IC 443, W44, W49B and W51C, with the Fermi Large Area Telescope. This detection provided direct evidence that cosmic-ray protons are (re-)accelerated in SNRs. Here, we present a comprehensive search for low-energy spectral breaks among 311 4FGL catalog sources located within 5 degrees from the Galactic plane. Using 8 years of data from the Fermi Large Area Telescope between 50 MeV and 1 GeV, we find and present the spectral characteristics of 56 sources with a spectral break confirmed by a thorough study of systematic uncertainty. Our population of sources includes 13 SNRs for which the proton-proton interaction is enhanced by the dense target material; the high-mass gamma-ray binary LS~I +61 303; the colliding wind binary eta Carinae; and the Cygnus star-forming region. This analysis better constrains the origin of the gamma-ray emission and enlarges our view to potential new cosmic-ray acceleration sites., Comment: Accepted for publication in The Astrophysical Journal

Recent experiments searching for sub-GeV/$c^2$ dark matter have observed event excesses close to their respective energy thresholds. Although specific to the individual technologies, the measured excess event rates have been consistently reported at or below event energies of a few-hundred eV, or with charges of a few electron-hole pairs. In the present work, we operated a 1-gram silicon SuperCDMS-HVeV detector at three voltages across the crystal (0 V, 60 V and 100 V). The 0 V data show an excess of events in the tens of eV region. Despite this event excess, we demonstrate the ability to set a competitive exclusion limit on the spin-independent dark matter--nucleon elastic scattering cross section for dark matter masses of $\mathcal{O}(100)$ MeV/$c^2$, enabled by operation of the detector at 0 V potential and achievement of a very low $\mathcal{O}(10)$ eV threshold for nuclear recoils. Comparing the data acquired at 0 V, 60 V and 100 V potentials across the crystal, we investigated possible sources of the unexpected events observed at low energy. The data indicate that the dominant contribution to the excess is consistent with a hypothesized luminescence from the printed circuit boards used in the detector holder.

After large galaxies merge, their central supermassive black holes are expected to form binary systems whose orbital motion generates a gravitational wave background (GWB) at nanohertz frequencies. Searches for this background utilize pulsar timing arrays, which perform long-term monitoring of millisecond pulsars (MSPs) at radio wavelengths. We use 12.5 years of Fermi Large Area Telescope data to form a gamma-ray pulsar timing array. Results from 35 bright gamma-ray pulsars place a 95\% credible limit on the GWB characteristic strain of $1.0\times10^{-14}$ at 1 yr$^{-1}$, which scales as the observing time span $t_{\mathrm{obs}}^{-13/6}$. This direct measurement provides an independent probe of the GWB while offering a check on radio noise models., Comment: 3 figures in the main text. 3 figures and 8 tables are in the supplementary material

Detailed numerical models of chromosphere and corona are required to understand the heating of the solar atmosphere. An accurate treatment of the solar chromosphere is complicated by the effects arising from Non Local Thermodynamic Equilibrium (NLTE) radiative transfer. A small number of strong, highly scattering lines dominate the cooling and heating in the chromosphere. Additionally, the recombination times of ionised hydrogen are longer than the dynamical timescales, requiring a non-equilibrium (NE) treatment of hydrogen ionisation. The MURaM code is extended to include the physical process required for accurate simulation of the solar chromosphere, as implemented in the Bifrost code. This includes a time-dependent treatment of hydrogen ionisation, a scattering multi-group radiation transfer scheme and approximations for NLTE radiative cooling. The inclusion of NE and NLTE physics has a large impact on the structure of the chromosphere; the NE treatment of hydrogen ionisation leads to a higher ionisation fraction and enhanced populations in the first excited state throughout cold inter-shock regions of the chromosphere. Additionally this prevents hydrogen ioniation from buffering energy fluctuations, leading to hotter shocks and cooler inter-shock regions. The hydrogen populations in the ground and first excited state are enhanced by $10^2-10^3$ in the upper chromosphere and up to $10^9$ near the transition region. Including the necessary NLTE physics leads to significant differences in chromospheric structure and dynamics. The thermodynamics and hydrogen populations calculated using the extended version of the MURaM code are consistent with previous non-equilibrium simulations. The electron number and temperature calculated using the non-equilibrium treatment of the chromosphere are required to accurately synthesise chromospheric spectral lines., Comment: 24 pages, 12 figures, to be published in A&A

Defect engineering is foundational to classical electronic device development and for emerging quantum devices. Here, we report on defect engineering of silicon single crystals with ion pulses from a laser accelerator with ion flux levels up to 10^22 ions/cm^2/s. Low energy ions from plasma expansion of the laser-foil target are implanted near the surface and then diffuse into silicon samples that were locally pre-heated by high energy ions. We observe low energy ion fluences of ~10^16 cm^-2, about four orders of magnitude higher than the fluence of high energy (MeV) ions. In the areas of highest energy deposition, silicon crystals exfoliate from single ion pulses. Color centers, predominantly W and G-centers, form directly in response to ion pulses without a subsequent annealing step. We find that the linewidth of G-centers increase in areas with high ion flux much more than the linewidth of W-centers, consistent with density functional theory calculations of their electronic structure. Laser ion acceleration generates aligned pulses of high and low energy ions that expand the parameter range for defect engineering and doping of semiconductors with tunable balances of ion flux, damage rates and local heating.

Graph Convolutional Networks (GCNs) is the state-of-the-art method for learning graph-structured data, and training large-scale GCNs requires distributed training across multiple accelerators such that each accelerator is able to hold a partitioned subgraph. However, distributed GCN training incurs prohibitive overhead of communicating node features and feature gradients among partitions for every GCN layer during each training iteration, limiting the achievable training efficiency and model scalability. To this end, we propose PipeGCN, a simple yet effective scheme that hides the communication overhead by pipelining inter-partition communication with intra-partition computation. It is non-trivial to pipeline for efficient GCN training, as communicated node features/gradients will become stale and thus can harm the convergence, negating the pipeline benefit. Notably, little is known regarding the convergence rate of GCN training with both stale features and stale feature gradients. This work not only provides a theoretical convergence analysis but also finds the convergence rate of PipeGCN to be close to that of the vanilla distributed GCN training without any staleness. Furthermore, we develop a smoothing method to further improve PipeGCN's convergence. Extensive experiments show that PipeGCN can largely boost the training throughput (1.7x~28.5x) while achieving the same accuracy as its vanilla counterpart and existing full-graph training methods. The code is available at https://github.com/RICE-EIC/PipeGCN., Comment: ICLR 2022

The SuperCDMS Collaboration is currently building SuperCDMS SNOLAB, a dark matter search focused on nucleon-coupled dark matter in the 1-5 GeV/c$^2$ mass range. Looking to the future, the Collaboration has developed a set of experience-based upgrade scenarios, as well as novel directions, to extend the search for dark matter using the SuperCDMS technology in the SNOLAB facility. The experienced-based scenarios are forecasted to probe many square decades of unexplored dark matter parameter space below 5 GeV/c$^2$, covering over 6 decades in mass: 1-100 eV/c$^2$ for dark photons and axion-like particles, 1-100 MeV/c$^2$ for dark-photon-coupled light dark matter, and 0.05-5 GeV/c$^2$ for nucleon-coupled dark matter. They will reach the neutrino fog in the 0.5-5 GeV/c$^2$ mass range and test a variety of benchmark models and sharp targets. The novel directions involve greater departures from current SuperCDMS technology but promise even greater reach in the long run, and their development must begin now for them to be available in a timely fashion. The experienced-based upgrade scenarios rely mainly on dramatic improvements in detector performance based on demonstrated scaling laws and reasonable extrapolations of current performance. Importantly, these improvements in detector performance obviate significant reductions in background levels beyond current expectations for the SuperCDMS SNOLAB experiment. Given that the dominant limiting backgrounds for SuperCDMS SNOLAB are cosmogenically created radioisotopes in the detectors, likely amenable only to isotopic purification and an underground detector life-cycle from before crystal growth to detector testing, the potential cost and time savings are enormous and the necessary improvements much easier to prototype., Comment: contribution to Snowmass 2021; v2 updated (assorted corrections and improvements to forecasts) October 2022; v3 updated (corrected SuperCDMS SNOLAB sensitivity curves in upgrade forecast plots in body of text) April 2023

Two photo-neutron sources, $^{88}$Y$^{9}$Be and $^{124}$Sb$^{9}$Be, have been used to investigate the ionization yield of nuclear recoils in the CDMSlite germanium detectors by the SuperCDMS collaboration. This work evaluates the yield for nuclear recoil energies between 1 keV and 7 keV at a temperature of $\sim$ 50 mK. We use a Geant4 simulation to model the neutron spectrum assuming a charge yield model that is a generalization of the standard Lindhard model and consists of two energy dependent parameters. We perform a likelihood analysis using the simulated neutron spectrum, modeled background, and experimental data to obtain the best fit values of the yield model. The ionization yield between recoil energies of 1 keV and 7 keV is shown to be significantly lower than predicted by the standard Lindhard model for germanium. There is a general lack of agreement among different experiments using a variety of techniques studying the low-energy range of the nuclear recoil yield, which is most critical for interpretation of direct dark matter searches. This suggests complexity in the physical process that many direct detection experiments use to model their primary signal detection mechanism and highlights the need for further studies to clarify underlying systematic effects that have not been well understood up to this point.

Cool main-sequence stars, such as the Sun, have magnetic fields which are generated by an internal dynamo mechanism. In the Sun, the dynamo mechanism produces a balance between the amounts of magnetic flux generated and lost over the Sun's 11-year activity cycle and it is visible in the Sun's different atmospheric layers using multi-wavelength observations. We used the same observational diagnostics, spanning several decades, to probe the emergence of magnetic flux on the two close by, active- and low-mass K dwarfs: 61 Cygni A and Epsilon Eridani. Our results show that 61 Cygni A follows the Solar dynamo with a regular cycle at all wavelengths, while Epsilon Eridani represents a more extreme level of the Solar dynamo, while also showing strong Solar-like characteristics. For the first time we show magnetic butterfly diagrams for stars other than the Sun. For the two K stars and the Sun, the rate at which the toroidal field is generated from surface poloidal field is similar to the rate at which toroidal flux is lost through flux emergence. This suggests that the surface field plays a crucial role in the dynamos of all three stars. Finally, for Epsilon Eridani, we show that the two chromospheric cycle periods, of ~3 and ~13 years, correspond to two superimposed magnetic cycles., Comment: 8 pages, 5 figures: Accepted by A&A

In fall 2020, Duke University's COVID-19 prevention strategy included risk reduction behaviors, frequent testing using pooled SARS-CoV-2 polymerase chain reaction testing, and contact tracing. Among 10,265 students who received testing 68,913 times, 84 had positive results. One half of infections were asymptomatic, and some had high viral loads. SARS-CoV-2 transmission was limited in this congregate setting by integration of prevention strategies that included identification of asymptomatic infections through frequent testing. Pooled testing reduced the need for resources while allowing high throughput with high sensitivity and rapid turnaround of results.

Coronal loops are the basic building block of the upper solar atmosphere. Comprehending how these are energized, structured, and evolve is key to understanding stellar coronae. Here we investigate how the energy to heat the loop is generated by photospheric magneto-convection, transported into the upper atmosphere, and how the internal structure of a coronal loop forms. In a 3D magnetohydrodynamics (MHD) model, we study an isolated coronal loop rooted with both footpoints in a shallow layer within the convection zone using the MURaM code. To resolve its internal structure, we limited the computational domain to a rectangular box containing a single coronal loop as a straightened magnetic flux tube. Field-aligned heat conduction, gray radiative transfer in the photosphere and chromosphere, and optically thin radiative losses in the corona were taken into account. The footpoints were allowed to interact self-consistently with the granulation surrounding them. The loop is heated by a Poynting flux that is self-consistently generated through small-scale motions within individual magnetic concentrations in the photosphere. Turbulence develops in the upper layers of the atmosphere as a response to the footpoint motions. We see little sign of heating by large-scale braiding of magnetic flux tubes from different photospheric concentrations at a given footpoint. The synthesized emission, as it would be observed by the Atmospheric Imaging Assembly or the X-ray Telescope, reveals transient bright strands that form in response to the heating events. Overall, our model roughly reproduces the properties and evolution of the plasma as observed within coronal loops. With this model we can build a coherent picture of how the energy flux to heat the upper atmosphere is generated near the solar surface and how this process drives and governs the heating and dynamics of a coronal loop., Comment: 19 pages, 16 figures

We propose a novel, structured pruning algorithm for neural networks -- the iterative, Sparse Structured Pruning algorithm, dubbed as i-SpaSP. Inspired by ideas from sparse signal recovery, i-SpaSP operates by iteratively identifying a larger set of important parameter groups (e.g., filters or neurons) within a network that contribute most to the residual between pruned and dense network output, then thresholding these groups based on a smaller, pre-defined pruning ratio. For both two-layer and multi-layer network architectures with ReLU activations, we show the error induced by pruning with i-SpaSP decays polynomially, where the degree of this polynomial becomes arbitrarily large based on the sparsity of the dense network's hidden representations. In our experiments, i-SpaSP is evaluated across a variety of datasets (i.e., MNIST, ImageNet, and XNLI) and architectures (i.e., feed forward networks, ResNet34, MobileNetV2, and BERT), where it is shown to discover high-performing sub-networks and improve upon the pruning efficiency of provable baseline methodologies by several orders of magnitude. Put simply, i-SpaSP is easy to implement with automatic differentiation, achieves strong empirical results, comes with theoretical convergence guarantees, and is efficient, thus distinguishing itself as one of the few computationally efficient, practical, and provable pruning algorithms., Comment: 29 pages, 4 figures, 4th Annual Conference on Learning for Dynamics and Control

A prototype Schwarzschild-Couder Telescope (pSCT) has been constructed at the Fred Lawrence Whipple Observatory as a candidate for the medium-sized telescopes of the Cherenkov Telescope Array Observatory (CTAO). CTAO is currently entering early construction phase of the project and once completed it will vastly improve very high energy gamma-ray detection component in multi-wavelength and multi-messenger observations due to significantly improved sensitivity, angular resolution and field of view comparing to the current generation of the ground-based gamma-ray observatories H.E.S.S., MAGIC and VERITAS. The pSCT uses a dual aspheric mirror design with a $9.7$ m primary mirror and $5.4$ m secondary mirror, both of which are segmented. The Schwarzschild-Couder (SC) optical system (OS) selected for the prototype telescope achieves wide field of view of $8$ degrees and simultaneously reduces the focal plane plate scale allowing an unprecedented compact ($0.78$m diameter) implementation of the high-resolution camera ($6$mm/ $0.067$deg per imaging pixel with $11,328$ pixels) based on the silicon photo-multipliers (SiPMs). The OS of the telescope is designed to eliminate spherical and comatic aberrations and minimize astigmatism to radically improve off-axis imaging and consequently angular resolution across all the field of view with respect to the conventional single-mirror telescopes. Fast and high imaging resolution OS of the pSCT comes with the challenging submillimeter-precision custom alignment system, which was successfully demonstrated with an on-axis point spread function (PSF) of $2.9$ arcmin prior to the first-light detection of the Crab Nebula in 2020. Ongoing and future commissioning activities are reported.

The Schwarzschild-Couder Telescope (SCT) is a medium-sized telescope technology proposed for the Cherenkov Telescope Array. It uses a novel dual-mirror optical design that removes comatic aberrations across its entire field of view. The SCT camera employs high-resolution silicon photomultiplier (SiPM) sensors with a pixel size of 4 arcminutes. A prototype SCT (pSCT) has been constructed at the Fred Lawrence Whipple Observatory in Arizona, USA. An observing campaign in 2020, with a partial camera of 1600 pixels (2.7 degrees by 2.7 degrees field of view) resulted in detection of the Crab Nebula at 8.6 sigma statistical significance. Work on the pSCT camera and optical system is ongoing to improve performance and prepare for an upcoming camera upgrade. The pSCT camera upgrade will replace the current camera modules with improved SiPMs and readout electronics and will expand the camera to its full design field of view of 8 degrees in diameter (11,328 pixels). The fully upgraded pSCT will enable next-generation very-high-energy gamma-ray astrophysics through excellent background rejection and angular resolution. In this presentation we describe first results from the successful operation of the pSCT and future plans., Comment: 9 pages, 3 figures, 2 tables, contribution to ICRC 2021, similar to 10.1016/j.astropartphys.2021.102562 (arXiv:2012.08448)

The Cherenkov Telescope Array (CTA) is the next-generation ground-based observatory for very-high-energy gamma-ray astronomy. An innovative 9.7 m aperture, dual-mirror Schwarzschild-Couder Telescope (SCT) design is a candidate design for CTA Medium-Sized Telescopes. A prototype SCT (pSCT) has been constructed at the Fred Lawrence Whipple Observatory in Arizona, USA. Its camera is currently partially instrumented with 1600 pixels covering a field of view of 2.7 degrees square. The small plate scale of the optical system allows densely packed silicon photomultipliers to be used, which combined with high-density trigger and waveform readout electronics enable the high-resolution camera. The camera's electronics are capable of imaging air shower development at a rate of one billion samples per second. We describe the commissioning and performance of the pSCT camera, including trigger and waveform readout performance, calibration, and absolute GPS time stamping. We also present the upgrade to the camera, which is currently underway. The upgrade will fully populate the focal plane, increasing the field of view to 8 degree diameter, and lower the front-end electronics noise, enabling a lower trigger threshold and improved reconstruction and background rejection., Comment: 8 pages, 5 figures, Proceedings of the 37th International Cosmic Ray Conference (ICRC 2021), Berlin, Germany

Neural network pruning is useful for discovering efficient, high-performing subnetworks within pre-trained, dense network architectures. More often than not, it involves a three-step process -- pre-training, pruning, and re-training -- that is computationally expensive, as the dense model must be fully pre-trained. While previous work has revealed through experiments the relationship between the amount of pre-training and the performance of the pruned network, a theoretical characterization of such dependency is still missing. Aiming to mathematically analyze the amount of dense network pre-training needed for a pruned network to perform well, we discover a simple theoretical bound in the number of gradient descent pre-training iterations on a two-layer, fully-connected network, beyond which pruning via greedy forward selection [61] yields a subnetwork that achieves good training error. Interestingly, this threshold is shown to be logarithmically dependent upon the size of the dataset, meaning that experiments with larger datasets require more pre-training for subnetworks obtained via pruning to perform well. Lastly, we empirically validate our theoretical results on a multi-layer perceptron trained on MNIST., Comment: 29 pages

By leveraging large amounts of product data collected across hundreds of live e-commerce websites, we construct 1000 unique classification tasks that share similarly-structured input data, comprised of both text and images. These classification tasks focus on learning the product hierarchy of different e-commerce websites, causing many of them to be correlated. Adopting a multi-modal transformer model, we solve these tasks in unison using multi-task learning (MTL). Extensive experiments are presented over an initial 100-task dataset to reveal best practices for "large-scale MTL" (i.e., MTL with more than 100 tasks). From these experiments, a final, unified methodology is derived, which is composed of both best practices and new proposals such as DyPa, a simple heuristic for automatically allocating task-specific parameters to tasks that could benefit from extra capacity. Using our large-scale MTL methodology, we successfully train a single model across all 1000 tasks in our dataset while using minimal task specific parameters, thereby showing that it is possible to extend several orders of magnitude beyond current efforts in MTL., Comment: 10 pages, 7 figures

We propose ResIST, a novel distributed training protocol for Residual Networks (ResNets). ResIST randomly decomposes a global ResNet into several shallow sub-ResNets that are trained independently in a distributed manner for several local iterations, before having their updates synchronized and aggregated into the global model. In the next round, new sub-ResNets are randomly generated and the process repeats until convergence. By construction, per iteration, ResIST communicates only a small portion of network parameters to each machine and never uses the full model during training. Thus, ResIST reduces the per-iteration communication, memory, and time requirements of ResNet training to only a fraction of the requirements of full-model training. In comparison to common protocols, like data-parallel training and data-parallel training with local SGD, ResIST yields a decrease in communication and compute requirements, while being competitive with respect to model performance., Comment: 26 pages, 8 figures, pre-print under review

The Large Area Telescope (LAT), the primary instrument for the Fermi Gamma-ray Space Telescope (Fermi) mission, is an imaging, wide field-of-view, high-energy gamma-ray telescope, covering the energy range from 30 MeV to more than 300 GeV. We describe the performance of the instrument at the 10-year milestone. LAT performance remains well within the specifications defined during the planning phase, validating the design choices and supporting the compelling case to extend the duration of the Fermi mission. The details provided here will be useful when designing the next generation of high-energy gamma-ray observatories., Comment: 60 pages, 28 figures. Published in ApJS