Source-linked AI summary

exoplanet: Gradient-based probabilistic inference for exoplanet data & other astronomical time series

Daniel Foreman-Mackey, Rodrigo Luger, Eric Agol, Thomas Barclay, Luke G. Bouma, Timothy D. Brandt, Ian Czekala, Trevor J. David, Jiayin Dong, Emily A. Gilbert, Tyler A. Gordon, Christina Hedges, Daniel R. Hey, Brett M. Morris, Adrian M. Price-Whelan, Arjun B. Savel

arXiv:2105.01994v2astro-ph.IMastro-ph.EP

TL;DR

Large and increasingly ambitious astronomical time-series datasets require flexible probabilistic models and scalable inference. exoplanet provides an ecosystem of differentiable, well-tested tools integrated with PyMC3, supporting exoplanet analysis and related astronomical time-series applications, with documented example use cases.

  • Problem

    Large astronomical surveys and increasingly ambitious physical models create computational and modeling challenges for exoplanet data analysis.

  • Method

    exoplanet provides a suite of libraries with specialized astronomical functions, efficient differentiable operations, and integration with PyMC3 and automatic differentiation.

  • Results

    The ecosystem includes automatically tested documentation and case studies demonstrating potential use cases for exoplanet software.

  • Takeaways & Limitations

    Differentiable model components support gradient-based inference and optimization for high-dimensional astronomical models.

Abstract

from arXiv · show

"exoplanet" is a toolkit for probabilistic modeling of astronomical time series data, with a focus on observations of exoplanets, using PyMC3 (Salvatier et al., 2016). PyMC3 is a flexible and high-performance model-building language and inference engine that scales well to problems with a large number of parameters. "exoplanet" extends PyMC3's modeling language to support many of the custom functions and probability distributions required when fitting exoplanet datasets or other astronomical time series. While it has been used for other applications, such as the study of stellar variability, the primary purpose of "exoplanet" is the characterization of exoplanets or multiple star systems using time-series photometry, astrometry, and/or radial velocity. In particular, the typical use case would be to use one or more of these datasets to place constraints on the physical and orbital parameters of the system, such as planet mass or orbital period, while simultaneously taking into account the effects of stellar variability.

Statement of need

Exoplanet targets increasingly ambitious time-domain astronomy questions using large survey datasets and flexible, computationally demanding models. Its ecosystem provides infrastructure for rigorous probabilistic inference with PyMC3.

  • Time-domain astronomy now offers huge survey datasets, including hundreds of thousands of observations per star from Kepler and TESS.
  • Research questions are becoming more ambitious in both the computational cost and flexibility of their physical models.
  • The exoplanet ecosystem is designed to enable rigorous probabilistic inference with large datasets and high-dimensional models.

The exoplanet software ecosystem

The exoplanet ecosystem combines a primary astronomy library with supporting packages for compiled operations and scalable time-series modeling. Together, these libraries provide related components for exoplanet data analysis.

  • The paper describes the full exoplanet software ecosystem, including the primary library and several related libraries.
  • exoplanet implements special functions for exoplanet data analysis, including orbital geometry, exoplanet-specific distributions, and exposure-time integrated transit light curves.
  • exoplanet-core provides efficient, tested, differentiable implementations of compiled exoplanet-specific operations, with numpy, jax, and PyMC3 support.
  • celerite2 supplies scalable Gaussian Process regression for time-series data and supports numpy, jax, and PyMC3.

Documentation & case studies

The ecosystem is supported by documentation and case-study websites whose tutorials and examples are automatically executed. Figure 1 illustrates fits to multi-survey photometry and multi-instrument radial-velocity data.

  • Documentation & case studies: The documentation tutorials are automatically executed on every GitHub push or pull request to check compatibility with the current code.
  • Documentation & case studies: The Case Studies website provides detailed example use cases and runs its examples weekly and after new exoplanet releases.
  • Documentation & case studies: Figure 1 shows a transiting-exoplanet light-curve fit combining Kepler and TESS observations.
  • Documentation & case studies: Figure 1 also shows phase-folded radial-velocity data from different observatories and instruments fit simultaneously with exoplanet.

Similar tools

Exoplanet is positioned among many astronomical inference tools but differs in emphasizing lower-level, fully differentiable model components integrated with PyMC3’s automatic-differentiation framework.

  • Similar tools: The paper lists related tools but does not provide a detailed comparison of their strengths and limitations.
  • Similar tools: Astronomical inference includes high-level fitting tools such as EXOFAST, radvel, juliet, exostriker, PYANETI, allesfitter, and orbitize.
  • Similar tools: These packages generally provide high-level interfaces for designing models and executing fits, whereas exoplanet is designed to be lower level.
  • Similar tools: Exoplanet supports differentiation of all model components and integrates with the aesara automatic-differentiation framework used by PyMC3.
  • Similar tools: Gradient support enables inference methods such as No-U-Turn Sampling and Automatic Differentiation Variational Inference for high-dimensional models.
Loading 2105.01994v2…