skip to main content

Title: An Interpretable Machine-learning Framework for Modeling High-resolution Spectroscopic Data*

Comparison of échelle spectra to synthetic models has become a computational statistics challenge, with over 10,000 individual spectral lines affecting a typical cool star échelle spectrum. Telluric artifacts, imperfect line lists, inexact continuum placement, and inflexible models frustrate the scientific promise of these information-rich data sets. Here we debut an interpretable machine-learning frameworkblaséthat addresses these and other challenges. The semiempirical approach can be viewed as “transfer learning”—first pretraining models on noise-free precomputed synthetic spectral models, then learning the corrections to line depths and widths from whole-spectrum fitting to an observed spectrum. The auto-differentiable model employs back-propagation, the fundamental algorithm empowering modern deep learning and neural networks. Here, however, the 40,000+ parameters symbolize physically interpretable line profile properties such as amplitude, width, location, and shape, plus radial velocity and rotational broadening. This hybrid data-/model-driven framework allows joint modeling of stellar and telluric lines simultaneously, a potentially transformative step forward for mitigating the deleterious telluric contamination in the near-infrared. Theblaséapproach acts as both a deconvolution tool and semiempirical model. The general-purpose scaffolding may be extensible to many scientific applications, including precision radial velocities, Doppler imaging, chemical abundances for Galactic archeology, line veiling, magnetic fields, and remote sensing. Its sparse-matrix architecture and GPU acceleration makeblaséfast. The open-source PyTorch-based codeblaseincludes tutorials, Application Programming Interface documentation, and more. We show how the tool fits into the existing Python spectroscopy ecosystem, demonstrate a range of astrophysical applications, and discuss limitations and future extensions.

more » « less
Award ID(s):
Author(s) / Creator(s):
Publisher / Repository:
DOI PREFIX: 10.3847
Date Published:
Journal Name:
The Astrophysical Journal
Medium: X Size: Article No. 200
["Article No. 200"]
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract

    Precise radial velocity (PRV) surveys are important for the search for Earth analogs around nearby bright stars, which induce a small stellar reflex motion with an RV amplitude of ∼10 cm s−1. Detecting such a small RV signal poses challenges to instrumentation, data analysis, and the precision of astrophysical models to mitigate stellar jitter. In this work, we investigate an important component in the PRV error budget—the spectral contamination from the Earth’s atmosphere (tellurics). We characterize the effects of telluric absorption on the RV precision and quantify its contribution to the RV error budget over time and across a wavelength range of 350 nm–2.5μm. We use simulated solar spectra with telluric contamination injected, and we extract the RVs using two commonly adopted algorithms: dividing out a telluric model before performing cross-correlation or forward modeling the observed spectrum incorporating a telluric model. We assume various degrees of cleanness in removing the tellurics. We conclude that the RV errors caused by telluric absorption can be suppressed to close to or even below 1–10 cm s−1in the blue optical region. At red through near-infrared wavelengths, however, the residuals of tellurics can induce an RV error on the meter-per-second level even under the most favorable assumptions for telluric removal, leading to significant systematic noise in the RV time series and periodograms. If the red-optical or near-infrared becomes critical in the mitigation of stellar activity, systematic errors from tellurics can be eliminated with a space mission such as EarthFinder.

    more » « less
  2. Abstract

    Spectroscopic studies of extreme-ionization galaxies (EIGs) are critical to our understanding of exotic systems throughout cosmic time. These EIGs exhibit spectral features requiring >54.42 eV photons: the energy needed to ionize helium into He2+fully and emit Heiirecombination lines. Spectroscopic studies of EIGs can probe exotic stellar populations or accretion onto intermediate-mass black holes (∼102–105M), which are the possibly key contributors to the reionization of the Universe. To facilitate the use of EIGs as probes of high-ionization systems, we focus on ratios constructed from several rest-frame UV/optical emission lines: [Oiii]λ5008, Hβ, [Neiii]λ3870, [Oii]λλ3727, 3729, and [Nev]λ3427. These lines probe the relative intensity at energies of 35.12, 13.62, 40.96, 13.62, and 97.12 eV, respectively, covering a wider range of ionization than traced by other common rest-frame UV/optical techniques. We use the ratios of these lines ([Nev]/[Neiii] ≡ Ne53, [Oiii]/Hβ, and [Neiii]/[Oii]), which are nearby in wavelength, mitigating the effects of dust attenuation and uncertainties in flux calibration. We make predictions from photoionization models constructed fromCloudythat use a broad range of stellar populations and black hole accretion models to explore the sensitivity of these line ratios to changes in the ionizing spectrum. We compare our models to observations from the Hubble Space Telescope and JWST of galaxies with strong high-ionization emission lines atz∼ 0,z∼ 2, and 5 <z< 8.5. We show that the Ne53 ratio can separate galaxies with ionization from “normal” stellar populations from those with active galactic nuclei and even “exotic” Population III models. We introduce new selection methods to identify galaxies with photoionization driven by Population III stars or intermediate-mass black hole accretion disks that could be identified in upcoming high-redshift spectroscopic surveys.

    more » « less
  3. Abstract

    We present a JWST/MIRI low-resolution mid-infrared (MIR) spectroscopic observation of the normal Type Ia supernova (SN Ia) SN 2021aefx at +323 days past rest-frameB-band maximum light. The spectrum ranges from 4 to 14μm and shows many unique qualities, including a flat-topped [Ariii] 8.991μm profile, a strongly tilted [Coiii] 11.888μm feature, and multiple stable Ni lines. These features provide critical information about the physics of the explosion. The observations are compared to synthetic spectra from detailed non–local thermodynamic equilibrium multidimensional models. The results of the best-fitting model are used to identify the components of the spectral blends and provide a quantitative comparison to the explosion physics. Emission line profiles and the presence of electron capture elements are used to constrain the mass of the exploding white dwarf (WD) and the chemical asymmetries in the ejecta. We show that the observations of SN 2021aefx are consistent with an off-center delayed detonation explosion of a near–Chandrasekhar mass (MCh) WD at a viewing angle of −30° relative to the point of the deflagration to detonation transition. From the strengths of the stable Ni lines, we determine that there is little to no mixing in the central regions of the ejecta. Based on both the presence of stable Ni and the Ar velocity distributions, we obtain a strict lower limit of 1.2Mfor the initial WD, implying that most sub-MChexplosions models are not viable models for SN 2021aefx. The analysis here shows the crucial importance of MIR spectra in distinguishing between explosion scenarios for SNe Ia.

    more » « less
  4. Abstract

    We introduce an end-to-end computational framework that allows for hyperparameter optimization using theDeepHyperlibrary, accelerated model training, and interpretable AI inference. The framework is based on state-of-the-art AI models includingCGCNN,PhysNet,SchNet,MPNN,MPNN-transformer, andTorchMD-NET. We employ these AI models along with the benchmarkQM9,hMOF, andMD17datasets to showcase how the models can predict user-specified material properties within modern computing environments. We demonstrate transferable applications in the modeling of small molecules, inorganic crystals and nanoporous metal organic frameworks with a unified, standalone framework. We have deployed and tested this framework in the ThetaGPU supercomputer at the Argonne Leadership Computing Facility, and in the Delta supercomputer at the National Center for Supercomputing Applications to provide researchers with modern tools to conduct accelerated AI-driven discovery in leadership-class computing environments. We release these digital assets as open source scientific software in GitLab, and ready-to-use Jupyter notebooks in Google Colab.

    more » « less

    The massive binary system formed by η Car and an unknown companion is a strong source at millimetre and submillimetre wavelengths. Close to the stars, continuum bremsstrahlung and radio recombination lines originate in the massive ionized wind of η Car and in several compact sources of high density plasma. Molecular lines are also detected at these wavelengths, some of them are seen in absorption towards the continuum emission region, many of them revealed by ALMA observations. However, because the ALMA atmospheric calibration is performed in a low spectral resolution mode, telluric lines can still be present in some high-resolution spectra of scientific products, which could lead to a false identification of molecules. In this work, we explore three different sets of ALMA archive data of η Car, including high resolution (0.065 arcsec × 0.043 arcsec) observations recently published by our group, to verify which of these absorption lines are real and discuss their origin. We conclude that some of them truly originate in clouds close to the binary system, while others are artefacts of a faulty elimination of telluric lines during ALMA calibration procedure. We found that these absorption lines are not present in the phase calibrators because they are much weaker than η Car, where the absorption line appears because the high intensity continuum enhances the small individual systematic calibration errors.

    more » « less