Last Updated: 20210607
AAS 238: Special Session 107
Unaccounted Uncertainties: The Role of Systematics in Astrophysics
Monday, 7 June 2021
Noon - 1:30pm EDT
This special session focuses on systematic uncertainties.
Systematic errors are present in all aspects of astrophysics, ranging from telescope calibration, multi-messenger astronomy, model accuracy, and numerical simulations.
We will explore the role they play, how specific cases may affect astrophysical inference, and discuss strategies to deal with them.
Many new modern technologies have been developed for dealing with data analysis issues in the age of `Big Data'.
These include tools and software designed for specific applications, such as for the Zwicky Transient Facility (ZTF), the Large Synoptic Survey Telescope (LSST), and the newly released data from Gaia, TESS and Kepler, or others focusing on specific issues in spatial-, spectral- and time-domains.
However, key challenges for the community remain: (1) to raise the awareness of astronomers to existing opportunities, and (2) to find the right software for application to particular problems.
The conjunction of this AAS with the Lab Astro Division meeting provides an extra impetus to discuss systematic uncertainties present in atomic data calculations and compilations. Thus, the first of the two talks focuses on atomic data errors. The second talk deals with efforts to handle cross-instrument systematics. The third half of the session will be devoted to a discussion of the issues from a statistical perspective.
- 12:05pm-12:30pm EDT: Stuart Loch (Auburn University), On Atomic Data Uncertainties
- Abstract: Analysis of astrophysical spectra places a great emphasis on statistical tools to ensure that the models being applied best fit the data which are observed. There are increasing efforts to model telescopes and instruments in ever more intensive ways to allow for uncertainties in the calibration and their subsequent effects on the spectra. However, while the spectroscopic diagnostics which can be extracted from these spectral are widely used powerful tools for identifying both the plasma state and composition, the contribution of uncertainties on the fundamental atomic data to such diagnostics has not been well understood. Work is described that outlines a method for assigning uncertainties on fundamental atomic structure and collision data, that can then be propagated through to uncertainties on spectroscopic diagnostics and plasma modeling codes. Structural uncertainties are inferred from available data by means of Bayesian analysis and their underlying distributions are interrogated through Markov chain Monte-Carlo sampling. The approach is designed to be general in nature and can be applied to a range of atomic processes, with examples given for dielectronic recombination and electron-impact excitation.
- Presentation Slides [.pptx]
- 12:30-12:55pm EDT: Yang Chen (Michigan University), Systematic Uncertainties in Multi-Telescope Observations
- Abstract: In this talk, I will cover published and recent work on astronomical instrument calibration with multiplicative shrinkage models. Calibration data are often obtained by observing several well-understood objects simultaneously with multiple instruments, such as satellites for measuring astronomical sources. Analyzing such data and obtaining proper concordance among the instruments is challenging when the physical source models are not well understood, when there are uncertainties in known physical quantities, or when data quality varies in ways that cannot be fully quantified. Furthermore, the number of model parameters increases with both the number of instruments and the number of sources. Thus, concordance of the instruments requires careful modeling of the mean signals, the intrinsic source differences, and measurement errors. We propose a log-Normal model and a more general log-t model that respect the multiplicative nature of the mean signals via a half-variance adjustment, yet permit imperfections in the mean modeling to be absorbed by residual variances. We demonstrate that our method provides helpful and practical guidance for astrophysicists when adjusting for disagreements among instruments. Extensions of the model to account for heterogeneity of noise levels and systematic correlation across instruments will be discussed if time permits.
- Presentation Slides [.pdf]
- 12:55-1:30pm EDT: Josh Speagle (University of Toronto), Discussant from a Statistics Perspective
- As a Statistics Discussant, I will provide statistical perspectives on systematics in astrophysics and help moderate the discussion involving the invited speakers and the audience.
- Presentation Slides [.pptx]
- To summarize the discussion, it seems that people feel confident that the approaches @Stuart Loch and @Yang Chen presented on today are useful first steps towards accounting for systematic uncertainties, and we should be looking towards the next steps of how to propagate them into downstream analysis in a computationally scalable and user-friendly manner. There also appear to be some good opportunities for using machine learning methods to both help model these uncertainties and improve scalability.
Vinay Kashyap (vkashyap @ cfa . harvard . edu)
Adam Foster (afoster @ cfa . harvard . edu)
Aneta Siemiginowska (asiemiginowska @ cfa . harvard . edu)
- 2021-apr-14: started page
- 2021-apr-21: small typo fixes
- 2021-jun-07: added speaker/discussant slides
CfA / CHASC