Publications

We consider when a sparse nonnegative matrix \(\mathbf{S}\) can be recovered, via an elementwise nonlinearity, from a real-valued matrix~ \(\mathbf{S}\) of significantly lower rank. Of particular interest is the setting where the positive elements of \( \mathbf{S}\) encode the similarities of nearby points on a low dimensional manifold. The recovery can then be posed as a problem in manifold learning---in this case, how to learn a norm-preserving and neighborhood-preserving mapping of high dimensional inputs into a lower dimensional space. We describe an algorithm for this problem based on a generalized low-rank decomposition of sparse matrices. This decomposition has the interesting property that it can be encoded by a neural network with one layer of rectified linear units; since the algorithm discovers this encoding, it can also be viewed as a layerwise primitive for deep learning. The algorithm regards the inputs \(\mathbf{x}_i|)\) and \(\mathbf{x}_j\)\) as similar whenever the cosine of the angle between them exceeds some threshold \(\tau\in(0,1)\). Given this threshold, the algorithm attempts to discover a mapping \(\mathbf{x}_i\mapsto\mathbf{y}_i\) by matching the elements of two sparse matrices; in particular, it seeks a mapping for which \(\mathbf{S}=\max(0,\mathbf{L})\), where \(S_{ij} = \max(0,\mathbf{x}_i\cdot\mathbf{x}_j - \tau\|\mathbf{x}_i\|\|\mathbf{x}_j\|)\) and \(L_{ij} = \mathbf{y}_i\cdot\mathbf{y}_j - \tau\|\mathbf{y}_i\|\|\mathbf{y}_j\|\). We apply the algorithm to data sets where vector magnitudes and small cosine distances have interpretable meanings (e.g., the brightness of an image, the similarity to other words). On these data sets, the algorithm is able to discover much lower dimensional representations that preserve these meanings

Many current approaches to machine learning in particle physics use generic architectures that require large numbers of parameters and disregard underlying physics principles, limiting their applicability as scientific modeling tools. In this work, we present a machine learning architecture that uses a set of inputs maximally reduced with respect to the full 6-dimensional Lorentz symmetry, and is fully permutation-equivariant throughout. We study the application of this network architecture to the standard task of top quark tagging and show that the resulting network outperforms all existing competitors despite much lower model complexity. In addition, we present a Lorentz-covariant variant of the same network applied to a 4-momentum regression task.

Observations show an almost ubiquitous presence of extra mixing in low-mass upper giant branch stars. The most commonly invoked explanation for this is thermohaline mixing. One-dimensional stellar evolution models include various prescriptions for thermohaline mixing, but the use of observational data directly to discriminate between thermohaline prescriptions has thus far been limited. Here, we propose a new framework to facilitate direct comparison: Using carbon-to-nitrogen measurements from the SDSS-IV APOGEE survey as a probe of mixing and a fluid parameter known as the reduced density ratio from one-dimensional stellar evolution programs, we compare the observed amount of extra mixing on the upper giant branch to predicted trends from three-dimensional fluid dynamics simulations. Using this method, we are able to empirically constrain how mixing efficiency should vary with the reduced density ratio. We find the observed amount of extra mixing is strongly correlated with the reduced density ratio and that trends between reduced density ratio and fundamental stellar parameters are robust across choices for modeling prescription. We show that stars with available mixing data tend to have relatively low density ratios, which should inform the regimes selected for future simulation efforts. Finally, we show that there is increased mixing at low reduced density ratios, which is consistent with current hydrodynamical models of thermohaline mixing. The introduction of this framework sets a new standard for theoretical modeling efforts, as validation for not only the amount of extra mixing, but trends between the degree of extra mixing and fundamental stellar parameters is now possible.

The core collapse of rapidly rotating massive ∼ 10M⊙ stars ("collapsars"), and the resulting formation of hyperaccreting black holes, comprise a leading model for the central engines of long-duration gamma-ray bursts (GRBs) and promising sources of r-process nucleosynthesis. Here, we explore the signatures of collapsars from progenitors with helium cores ≳ 130M⊙ above the pair-instability mass gap. While the rapid collapse to a black hole likely precludes prompt explosions in these systems, we demonstrate that disk outflows can generate a large quantity (up to ≳ 50M⊙) of ejecta, comprised of ≳ 5–10M⊙ in r-process elements and ∼ 0.1–1M⊙ of 56Ni, expanding at velocities ∼0.1 c. Radioactive heating of the disk wind ejecta powers an optical/IR transient, with a characteristic luminosity ∼ 1042 erg s−1 and a spectral peak in the near-IR (due to the high optical/UV opacities of lanthanide elements), similar to kilonovae from neutron star mergers, but with longer durations ≳1 month. These "super-kilonovae" (superKNe) herald the birth of massive black holes ≳ 60M⊙, which—as a result of disk wind mass loss—can populate the pair-instability mass gap "from above," and could potentially create the binary components of GW190521. SuperKNe could be discovered via wide-field surveys, such as those planned with the Roman Space Telescope, or via late-time IR follow-up observations of extremely energetic GRBs. Multiband gravitational waves of ∼ 0.1–50 Hz from nonaxisymmetric instabilities in self-gravitating massive collapsar disks are potentially detectable by proposed observatories out to hundreds of Mpc; in contrast to the "chirp" from binary mergers, the collapsar gravitational-wave signal decreases in frequency as the disk radius grows ("sad trombone").

Here, we report the discovery of a kilonova associated with the nearby (350 Mpc) minute-duration GRB 211211A. In tandem with deep optical limits that rule out the presence of an accompanying supernova to MI>−13 mag at 17.7 days post-burst, the identification of a kilonova confirms that this burst's progenitor was a compact object merger. While the spectrally softer tail in GRB 211211A's gamma-ray light curve is reminiscent of previous extended emission short GRBs (EE-SGRBs), its prompt, bright spikes last ≳12 s, separating it from past EE-SGRBs. GRB 211211A's kilonova has a similar luminosity, duration and color to AT2017gfo, the kilonova found in association with the gravitational wave (GW)-detected binary neutron star (BNS) merger GW170817. We find that the merger ejected ≈0.04M⊙ of r-process-rich material, and is consistent with the merger of two neutron stars (NSs) with masses close to the canonical 1.4M⊙. This discovery implies that GRBs with long, complex light curves can be spawned from compact object merger events and that a population of kilonovae following GRBs with durations ≫2 s should be accounted for in calculations of the NS merger r-process contribution and rate. At 350 Mpc, the current network of GW interferometers at design sensitivity would have detected the merger precipitating GRB 211211A, had it been operating at the time of the event. Further searches for GW signals coincident with long GRBs are therefore a promising route for future multi-messenger astronomy.

For decades, gamma-ray bursts (GRBs) have been broadly divided into `long'- and `short'-duration bursts, lasting more or less than 2s, respectively. However, this dichotomy does not map perfectly to the two progenitor channels that are known to produce GRBs -- the merger of compact objects (merger-GRBs) or the collapse of massive stars (collapsar-GRBs). In particular, the merger-GRBs population may also include bursts with a short, hard ≲2s spike and subsequent longer, softer extended emission (EE). The recent discovery of a kilonova -- the radioactive glow of heavy elements made in neutron star mergers -- in the 50s-duration GRB 211211A further demonstrates that mergers can drive long, complex GRBs that mimic the collapsar population. Here we present a detailed temporal and spectral analysis of the high-energy emission of GRB 211211A. We demonstrate that the emission has a purely synchrotron origin, with both the peak and cooling frequencies moving through the γ-ray band down to the X-rays, and that the rapidly-evolving spectrum drives the EE signature at late times. The identification of such spectral evolution in a merger-GRB opens avenues for diagnostics of the progenitor type.

The distance to the Galactic center R0 is a fundamental parameter for understanding the Milky Way, because all observations of our Galaxy are made from our heliocentric reference point. The uncertainty in R0 limits our knowledge of many aspects of the Milky Way, including its total mass and the relative mass of its major components, and any orbital parameters of stars employed in chemo-dynamical analyses. While measurements of R0 have been improving over a century, measurements in the past few years from a variety of methods still find a wide range of R0 being somewhere within 8.0 to 8.5kpc. The most precise measurements to date have to assume that Sgr A∗ is at rest at the Galactic center, which may not be the case. In this paper, we use maps of the kinematics of stars in the Galactic bar derived from APOGEE DR17 and Gaia EDR3 data augmented with spectro-photometric distances from the \texttt{astroNN} neural-network method. These maps clearly display the minimum in the rotational velocity vT and the quadrupolar signature in radial velocity vR expected for stars orbiting in a bar. From the minimum in vT, we measure R0=8.23±0.12kpc. We validate our measurement using realistic N-body simulations of the Milky Way. We further measure the pattern speed of the bar to be Ωbar=40.08±1.78kms−1kpc−1. Because the bar forms out of the disk, its center is manifestly the barycenter of the bar+disc system and our measurement is therefore the most robust and accurate measurement of R0 to date.

Recent experiments have shown that the nematode {\it T. aceti} can assemble into collectively undulating groups at the edge of fluid drops. This coordinated state consists of metachronal waves and drives fluid circulation inside the drop. We find that the circulation velocity is about 2 mm/s and nearly half the speed of the metachronal wave. We develop a quasi two-dimensional hydrodynamics model using the Stokes flow approximation. The periodic motion of the nematodes constitute our moving boundary condition that drives the flow. Our model suggests that large amplitude excursions of the nematodes tails produce the fluid circulation. We discuss the constraints on containers that would enhance fluid motion, which could be used in the future design of on demand flow generating systems.

The rupture potential of an atherosclerotic plaque is dependent on both the plaque's composition and the shear stresses it encounters from blood flow. Because plaques move and deform throughout the cardiac cycle, resulting in changes to plaque position and shape as well as to the encountered shear stresses, concurrent imaging of both risk factors over time is required to accurately predict plaque vulnerability. To evaluate the potential to achieve as much, multi-angle plane wave (PW) ARFI and least-squares vector Doppler data were acquired in a calibrated flow phantom with channels of 4–8 mm diameters and flow rates of 100–600 ml/min. The wall shear stress (WSS) was measured to within 15% of the ground-truth analytical solutions. The same methods were then implemented in an excised human cadaveric carotid with a x% stenotic plaque. ARFI VoA detected plaque regions of calcium and intraplaque hemorrhage that were validated by spatially-matched histology. Concurrent vector Doppler yielded a peak WSS of 5.2 Pa on the plaque shoulder, which was consistent with the 6.4 Pa WSS predicted by an immersed interface fluid-solid interation (FSI) model developed using the specific geometry of the examined cadaveric carotid. Overall our results demonstrate the feasibility of concurrent imaging of carotid plaque composition by ARFI VoA, vector flow, and WSS to better assess stroke risk.

We examine the ground-state phase diagram and thermal phase transitions in a plaquettized fully frustrated bilayer spin-1/2 Heisenberg model. Based on a combined analysis from sign-problem free quantum Monte Carlo simulations, perturbation theory and free-energy arguments, we identify a first-order quantum phase transition line that separates two competing quantum-disordered ground states with dominant singlet formations on inter-layer dimers and plaquettes, respectively. At finite temperatures, this line extends to form a wall of first-order thermal transitions, which terminates in a line of thermal critical points. From a perturbative approach in terms of an effective Ising model description, we identify a quadratic suppression of the critical temperature scale in the strongly plaquettized region. Based on free-energy arguments we furthermore obtain the full phase boundary of the low-temperature dimer-singlet regime, which agrees well with the quantum Monte Carlo data.