Publications

Despite over three hundred years of effort, no solutions exist for predicting when a general planetary configuration will become unstable. We introduce a deep learning architecture to push forward this problem for compact systems. While current machine learning algorithms in this area rely on scientist-derived instability metrics, our new technique learns its own metrics from scratch, enabled by a novel internal structure inspired from dynamics theory. Our Bayesian neural network model can accurately predict not only if, but also when a compact planetary system with three or more planets will go unstable. Our model, trained directly from short N-body time series of raw orbital elements, is more than two orders of magnitude more accurate at predicting instability times than analytical estimators, while also reducing the bias of existing machine learning algorithms by nearly a factor of three. Despite being trained on compact resonant and near-resonant three-planet configurations, the model demonstrates robust generalization to both non-resonant and higher multiplicity configurations, in the latter case outperforming models fit to that specific set of integrations. The model computes instability estimates up to five orders of magnitude faster than a numerical integrator, and unlike previous efforts provides confidence intervals on its predictions. Our inference model is publicly available in the SPOCK package, with training code open-sourced.

Despite over three hundred years of effort, no solutions exist for predicting when a general planetary configuration will become unstable. We introduce a deep learning architecture to push forward this problem for compact systems. While current machine learning algorithms in this area rely on scientist-derived instability metrics, our new technique learns its own metrics from scratch, enabled by a novel internal structure inspired from dynamics theory. Our Bayesian neural network model can accurately predict not only if, but also when a compact planetary system with three or more planets will go unstable. Our model, trained directly from short N-body time series of raw orbital elements, is more than two orders of magnitude more accurate at predicting instability times than analytical estimators, while also reducing the bias of existing machine learning algorithms by nearly a factor of three. Despite being trained on compact resonant and near-resonant three-planet configurations, the model demonstrates robust generalization to both non-resonant and higher multiplicity configurations, in the latter case outperforming models fit to that specific set of integrations. The model computes instability estimates up to five orders of magnitude faster than a numerical integrator, and unlike previous efforts provides confidence intervals on its predictions. Our inference model is publicly available in the SPOCK package, with training code open-sourced.

Indications of disequilibrium throughout the Milky Way (MW) highlight the need for compact, flexible, non-parametric descriptions of phase–space distributions of galaxies. We present a new representation of the current dark matter (DM) distribution and potential derived from N-body simulations of the MW and Large Magellanic Cloud (LMC) system using basis function expansions (BFEs). We incorporate methods to maximize the physical signal in the representation. As a result, the simulations of 108 DM particles representing the distorted MW(MW+LMC) system can be described by ∼236(2067) coefficients. We find that the LMC induces asymmetric perturbations (odd l, m) to the MW’s halo, which are inconsistent with oblate, prolate, or triaxial halos. Furthermore, the energy in high order even modes (l, m > 2) is similar to average triaxial halos found in cosmological simulations. As such, the response of the MW’s halo to the LMC must be accounted for in order to recover the imprints of its assembly history. The LMC causes the outer halo (>30 kpc) to shift from the disk center of mass (COM) by ∼15–25 kpc at present day, manifesting as a dipole in the BFE and in the radial velocities of halo stars. The shift depends on the LMC’s infall mass, the distortion of the LMC’s halo and the MW halo response.Within 30 kpc, halo tracers are expected to orbit the COM of the MW’s disk, regardless of LMC infall mass. The LMC’s halo is also distorted by MW tides; we discuss the implications for its mass loss and the subsequent effects on current Magellanic satellites.

The long-term goal of this work is the development of high-fidelity simulation tools for dispersive tsunami propagation. A dispersive model is especially important for short wavelength phenomena such as an asteroid impact into the ocean, and is also important in modeling other events where the simpler shallow water equations are insufficient. Adaptive simulations are crucial to bridge the scales from deep ocean to inundation, but have difficulties with the implicit system of equations that results from dispersive models. We propose a fractional step scheme that advances the solution on separate patches with different spatial resolutions and time steps. We show a simulation with 7 levels of adaptive meshes and onshore inundation resulting from a simulated asteroid impact off the coast of Washington. Finally, we discuss a number of open research questions that need to be resolved for high quality simulations.

Stacking is a widely used model averaging technique that asymptotically yields optimal predictions among linear averages. We show that stacking is most effective when model predictive performance is heterogeneous in inputs, and we can further improve the stacked mixture with a hierarchical model. We generalize stacking to Bayesian hierarchical stacking. The model weights are varying as a function of data, partially-pooled, and inferred using Bayesian inference. We further incorporate discrete and continuous inputs, other structured priors, and time series and longitudinal data. To verify the performance gain of the proposed method, we derive theory bounds, and demonstrate on several applied problems.

With the advent of large spectroscopic surveys the amount of high quality chemodynamical data in the Milky Way (MW) increased tremendously. Accurately and correctly capturing and explaining the detailed features in the high-quality observational data is notoriously difficult for state-of-the-art numerical models. In order to keep up with the quantity and quality of observational data sets, improved prescriptions for galactic chemical evolution need to be incorporated into the simulations. Here we present a new, flexible, time-resolved chemical enrichment model for cosmological simulations. Our model allows us to easily change a number of stellar physics parameters such as the shape of the initial mass function (IMF), stellar lifetimes, chemical yields, or SN Ia delay times. We implement our model into the Gasoline2 code and perform a series of cosmological simulations varying a number of key parameters, foremost evaluating different stellar yield sets for massive stars from the literature. We find that total metallicity, total iron abundance, and gas phase oxygen abundance are robust predictions from different yield sets and in agreement with observational relations. On the other hand, individual element abundances, especially alpha-elements show significant differences across different yield sets and none of our models can simultaneously match constraints on the dwarf and MW mass scale. This offers a unique way of observationally constraining model parameters. For MW mass galaxies we find for most yield tables tested in this work a bimodality in the [α/Fe] versus [Fe/H] plane of rather low intrinsic scatter potentially in tension with the observed abundance scatter.

Aims. Spectroscopic surveys have by now collectively observed tens of thousands of stars in the bulge of our Galaxy. However, each of these surveys had unique observing and data processing strategies that led to distinct stellar parameter and abundance scales. Because of this, stellar samples from different surveys cannot be directly combined.

Methods. Here we use the data-driven method, The Cannon, to bring 21 000 stars from the ARGOS bulge survey, including 10 000 red clump stars, onto the parameter and abundance scales of the cross-Galactic survey, APOGEE, obtaining rms precisions of 0.10 dex, 0.07 dex, 74 K, and 0.18 dex for [Fe/H], [Mg/Fe], Teff, and log(g), respectively. The re-calibrated ARGOS survey – which we refer to as the A2A survey – is combined with the APOGEE survey to investigate the abundance structure of the Galactic bulge.

Results. We find X-shaped [Fe/H] and [Mg/Fe] distributions in the bulge that are more pinched than the bulge density, a signature of its disk origin. The mean abundance along the major axis of the bar varies such that the stars are more [Fe/H]-poor and [Mg/Fe]-rich near the Galactic centre than in the outer bulge and the long bar region. The vertical [Fe/H] and [Mg/Fe] gradients vary between the inner bulge and the long bar, with the inner bulge showing a flattening near the plane that is absent in the long bar. The [Fe/H] − [Mg/Fe] distribution shows two main maxima, an ‘[Fe/H]-poor [Mg/Fe]- rich’ maximum and an ‘[Fe/H]-rich [Mg/Fe]-poor’ maximum, that vary in strength with position in the bulge. In particular, the outer long bar close to the Galactic plane is dominated by super-solar [Fe/H], [Mg/Fe]-normal stars. Stars composing the [Fe/H]-rich maximum show little kinematic dependence on [Fe/H], but for lower [Fe/H] the rotation and dispersion of the bulge increase slowly. Stars with [Fe/H] < −1 dex have a very different kinematic structure than stars with higher [Fe/H]. Conclusions. Comparing with recent models for the Galactic boxy-peanut bulge, the abundance gradients and distribution, and the relation between [Fe/H] and kinematics suggests that the stars comprising each maximum have separate disk origins with the ‘[Fe/H]-poor [Mg/Fe]-rich’ stars originating from a thicker disk than the ‘[Fe/H]-rich [Mg/Fe]-poor’ stars.

We use the data-driven method, The Cannon, to bring 21,000 stars from the ARGOS bulge survey, including 10,000 red clump stars, onto the parameter and abundance scales of the cross-Galactic survey, APOGEE, obtaining rms precisions of 0.10 dex, 0.07 dex, 74 K, and 0.18 dex for [Fe/H], [Mg/Fe], Teff, and log(g), respectively. The re-calibrated ARGOS survey - which we refer to as the A2A survey - is combined with the APOGEE survey to investigate the abundance structure of the Galactic bulge. We find X-shaped [Fe/H] and [Mg/Fe] distributions in the bulge that are more pinched than the bulge density, a signature of its disk origin. The mean abundance along the major axis of the bar varies such that the stars are more [Fe/H]-poor and [Mg/Fe]-rich near the Galactic center than in the long bar/outer bulge region. The vertical [Fe/H] and [Mg/Fe] gradients vary between the inner bulge and long bar with the inner bulge showing a flattening near the plane that is absent in the long bar. The [Fe/H]-[Mg/Fe] distribution shows two main maxima, an ``[Fe/H]-poor [Mg/Fe]- rich'' maximum and an ``[Fe/H]-rich [Mg/Fe]-poor'' maximum, that vary in strength with position in the bulge. In particular, the outer long bar close to the Galactic plane is dominated by super-solar [Fe/H], [Mg/Fe]-normal stars. Stars composing the [Fe/H]-rich maximum show little kinematic dependence on [Fe/H], but for lower [Fe/H] the rotation and dispersion of the bulge increase slowly. Stars with [Fe/H]<-1 dex have a very different kinematic structure than stars with higher [Fe/H]. Comparing with recent models for the Galactic boxy-peanut bulge, the abundance gradients and distribution, and the relation between [Fe/H] and kinematics suggest that the stars comprising each maximum have separate disk origins with the ``[Fe/H]-poor [Mg/Fe]-rich'' stars originating from a thicker disk than the ``[Fe/H]-rich [Mg/Fe]-poor'' stars.

One of the hallmarks of the cerebral cortex is the extreme diversity of interneurons. The two largest subtypes of cortical interneurons, parvalbumin- and somatostatin-positive cells, are morphologically and functionally distinct in adulthood but arise from common lineages within the medial ganglionic eminence.This makes them an attractive model for studying the generation of cell diversity. Here we examine how developmental changes in transcription and chromatin structure enable these cells to acquire distinct identities in the mouse cortex. Generic interneuron features are first detected upon cell cycle exit through the opening of chromatin at distal elements. By constructing cell-type-specific gene regulatory networks, we observed that parvalbumin- and somatostatin-positive cells initiate distinct programs upon settling within the cortex. We used these networks to model the differential transcriptional requirement of a shared regulator, Mef2c, and confirmed the accuracy of our predictions through experimental loss-of-function experiments. We therefore reveal how a common molecular program diverges to enable these neuronal subtypes to acquire highly specialized properties by adulthood. Our methods provide a framework for examining the emergence of cellular diversity, as well as for quantifying and predicting the effect of candidate genes on cell-type-specific development.