The Hubble Catalog of Variables (HCV)

K. V. Sokolovsky, A. Z. Bonanos, P. Gavras, Ming Yang, D. Hatzidimitriou, M. I. Moretti, A. Karampelas, I. Bellas-Velidis, Z. T. Spetsieri, E. Pouliasis, I. Georgantopoulos, V. Charmandaris, K. Tsinganos, Nikolaos Laskaris, George Kakaletris, A. Nota, D. J. Lennon, C. Arviset, Bradley C. Whitmore, Tamás Budavári, R. A. Downes, Stephen H. Lubow, A. Rest, Louis-Gregory Strolger, R. L. WhiteProceedings of the International Astronomical Union 2017 · DOI · arXiv

TL;DR

The Hubble Catalog of Variables (HCV) is the first comprehensive effort to systematically identify variable light sources across the vast Hubble Source Catalog (HSC). By analyzing nearly 2 million sources, researchers identified approximately 52,000 candidate variables, providing a massive new resource for studying stellar evolution, galactic structure, and active galactic nuclei.

What Problem Does This Paper Address?

For decades, the Hubble Space Telescope (HST) has captured high-resolution images of the universe, but these observations are often fragmented across thousands of individual research programs. While the Hubble Source Catalog (HSC) combined these into a master list of objects, it didn't inherently highlight which ones change over time.

Detecting variability in the Hubble archive is challenging because the data is "sparse"—many objects were only visited a handful of times. Standard methods like period-searching (used for pulsars or eclipsing binaries) fail when you only have five or ten data points. The HCV project aims to create a reliable pipeline to distinguish true astrophysical variability from instrumental noise, cosmic rays, and processing artifacts across the entire HST archive.

Methods

The researchers developed a multi-stage pipeline to process 1.85 million sources that had at least five visits from the same instrument and filter.

  • Pre-processing & Quality Control: They calculated a "synthetic error" for every data point, combining standard photometric error with metadata like the "concentration index" (to detect blending) and aperture differences. Sources with fewer than five good points were discarded.
  • Local Corrections: To account for small-scale sensitivity variations across Hubble’s detectors, they applied local magnitude zero-point corrections by comparing a source to its neighbors within a 20-arcsecond radius.
  • The Variability Statistic: Instead of standard deviation, they used Median Absolute Deviation (MAD). MAD is far more robust against outliers (like a single bad pixel or a cosmic ray hit). They established a magnitude-dependent threshold, flagging sources that sat at least 5$\sigma$ above the median MAD for their brightness level.
  • Refinement: They applied a reduced $\chi^2$ cut ($\chi^2_{red} > 3$) to ensure the observed scatter was significantly larger than the expected measurement errors.

Datasets

  • Hubble Source Catalog (HSC) v2: The foundational dataset, containing sources from the WFPC2, ACS, and WFC3 instruments.
  • Ancillary Catalogs: The team used the Guide Star Catalog (GSC), Pan-STARRS, 2MASS, and SDSS to cross-match and improve astrometric (positional) accuracy.

Key Insights and Ideas

  • Robustness over Complexity: By choosing MAD as their primary statistic, the team prioritized reliability over the ability to define the type of variability. This allowed them to flag candidates with as few as five observations, where a Fourier-based period search would be impossible.
  • Magnitude-Dependent Thresholds: Variability detection isn't "one size fits all." Because noise increases as objects get fainter, the threshold for what constitutes "significant" scatter must be calculated as a function of the object's magnitude (as seen in Figure 1 of the paper).
  • The "Human-in-the-Loop" Requirement: Automated pipelines are still imperfect. Visual inspection of multi-filter candidates revealed that ~30% were false positives caused by blending or imaging artifacts, emphasizing that "Big Data" in astronomy still requires expert validation.

Key Results

  • Candidate Yield: Out of 1.85 million eligible sources, 52,000 (2.8%) were identified as candidate variables.
  • Multi-Filter Validation: 7,800 of these candidates showed variability in more than one filter, significantly increasing the confidence that the variation is physical rather than instrumental.
  • Depth: The catalog reaches incredibly faint magnitudes, ranging from 15 to 27 mag (with a median of 22 mag), allowing for the study of variables in distant galaxies that are inaccessible to ground-based surveys.
  • Reliability: Visual inspection suggests a ~70% success rate for candidates detected in multiple filters.

Limitations and Future Work

  • Artifacts and Blending: About 30% of the initial candidates are likely "corrupted" by blending with nearby stars or processing anomalies. Future versions of the HSC aim to improve the "white-light" image alignment to reduce these false positives.
  • Sparse Sampling: Because many sources have very few observations, the HCV can identify that an object is variable, but often cannot determine how it varies (e.g., its period or light-curve shape).
  • Future Releases: The authors planned to release both a "clean" sample (high confidence) and an "extended" sample (automated selection only) to the community, which has since become a vital tool for time-domain astronomy.

Summary updated

Abstract

Abstract The Hubble Source Catalog (HSC) combines lists of sources detected on images obtained with the WFPC2, ACS and WFC3 instruments aboard the Hubble Space Telescope (HST) and now available in the Hubble Legacy Archive. The catalogue contains time-domain information for about two million of its sources detected using the same instrument and filter on at least five HST visits. The Hubble Catalog of Variables (HCV) aims to identify HSC sources showing significant brightness variations. A magnitude-dependent threshold in the median absolute deviation of photometric measurements (an outlier-resistant measure of light-curve scatter) is adopted as the variability detection statistic. It is supplemented with a cut in χ red 2 that removes sources with large photometric errors. A pre-processing procedure involving bad image identification, outlier rejection and computation of local magnitude zero-point corrections is applied to the HSC light-curves before computing the variability detection statistics. About 52 000 HSC sources have been identified as candidate variables, among which 7,800 show variability in more than one filter. Visual inspection suggests that ∼70% of the candidates detected in multiple filters are true variables, while the remaining ∼30% are sources with aperture photometry corrupted by blending, imaging artefacts or image processing anomalies. The candidate variables have AB magnitudes in the range 15–27 m , with a median of 22 m . Among them are the stars in our own and nearby galaxies, and active galactic nuclei.
BibTeX
@inproceedings{sokolovsky2017hubble,
  author = {K. V. Sokolovsky and A. Z. Bonanos and P. Gavras and Ming Yang and D. Hatzidimitriou and M. I. Moretti and A. Karampelas and I. Bellas-Velidis and Z. T. Spetsieri and E. Pouliasis and I. Georgantopoulos and V. Charmandaris and K. Tsinganos and Nikolaos Laskaris and George Kakaletris and A. Nota and D. J. Lennon and C. Arviset and Bradley C. Whitmore and Tamás Budavári and R. A. Downes and Stephen H. Lubow and A. Rest and Louis-Gregory Strolger and R. L. White},
  title = {{The Hubble Catalog of Variables (HCV)}},
  booktitle = {Proceedings of the International Astronomical Union},
  year = {2017},
  doi = {10.1017/s1743921318002296},
}
Figure