51�������

Abstract:

Computer simulations are invaluable tools for scientific discovery. However, accurate simulations are often slow to execute, which limits their applicability to extensive parameter exploration, large-scale data analysis, and uncertainty quantification. A promising route to accelerate simulations by building fast emulators with machine learning requires large training datasets, which can be prohibitively expensive to obtain with slow simulations. Here we present a method based on neural architecture search to build accurate emulators even with a limited number of training data. The method successfully emulates simulations in 10 scientific cases including astrophysics, climate science, biogeochemistry, high energy density physics, fusion energy, and seismology, using the same super-architecture, algorithm, and hyperparameters. Our approach also inherently provides emulator uncertainty estimation, adding further confidence in their use. We anticipate this work will accelerate research involving expensive simulations, allow more extensive parameters exploration, and enable new, previously unfeasible computational discovery.

Abstract:

We describe the survey design, calibration, commissioning, and emission-line detection algorithms for the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX). The goal of HETDEX is to measure the redshifts of over a million Lyα emitting galaxies between 1.88 < z < 3.52, in a 540 deg2 area encompassing a co-moving volume of 10.9 Gpc3. No pre-selection of targets is involved; instead the HETDEX measurements are accomplished via a spectroscopic survey using a suite of wide-field integral field units distributed over the focal plane of the telescope. This survey measures the Hubble expansion parameter and angular diameter distance, with a final expected accuracy of better than 1%. We detail the project’s observational strategy, reduction pipeline, source detection, and catalog generation, and present initial results for science verification in the COSMOS, Extended Groth Strip, and GOODS-N fields. We demonstrate that our data reach the required specifications in throughput, astrometric accuracy, flux limit, and object detection, with the end products being a catalog of emission-line sources, their object classifications, and flux-calibrated spectra.

Abstract:

ABSTRACT We introduce and apply a new approach to probe the response of galactic stellar haloes to the interplay between cosmological merger histories and galaxy formation physics. We perform dark matter-only, zoomed simulations of two Milky Way-mass hosts and make targeted, controlled changes to their cosmological histories using the genetic modification technique. Populating each history’s stellar halo with a semi-empirical, particle tagging approach then enables a controlled study, with all instances converging to the same large-scale structure, dynamical and stellar mass at z = 0 as their reference. These related merger scenarios alone generate an extended spread in stellar halo mass fractions (1.5 dex) comparable to the observed population, with the largest scatter achieved by growing late (z ≤ 1) major mergers that spread out existing stars to create massive, in-situ dominated stellar haloes. Increasing a last major merger at z ∼ 2 brings more accreted stars into the inner regions, resulting in smaller scatter in the outskirts which are predominantly built by subsequent minor events. Exploiting the flexibility of our semi-empirical approach, we show that the diversity of stellar halo masses across scenarios is reduced by allowing shallower slopes in the stellar mass–halo mass relation for dwarf galaxies, while it remains conserved when central stars are born with hotter kinematics across cosmic time. The merger-dependent diversity of stellar haloes thus responds distinctly to assumptions in modelling the central and dwarf galaxies respectively, opening exciting prospects to constrain star formation and feedback at different galactic mass-scales with the coming generation of deep, photometric observatories.