PSComet: Pan-STARRS1 Comet-Search System

• PSComet is a detection system that identifies weak cometary activity by comparing the radial PSFs of moving objects to nearby field stars.
• It integrates automated screening with human vetting to distinguish between genuine low-level activity and common imaging artifacts.
• The system employs complementary methods—including linear-PSF analysis, extended-source photometry, and crowd-sourcing—to enhance the detection efficiency of main-belt comets.

PSComet is the Pan-STARRS1 comet-search system used to identify cometary activity among moving objects by comparing the point spread functions (PSFs) of moving detections to those of nearby field stars and flagging objects that show anomalously large radial PSF widths. In the Pan-STARRS1 survey analysis of 760475 observations of 333026 main-belt objects obtained between 2012 May 20 and 2013 November 9, this procedure operated within the broader search for main-belt comets (MBCs), a class whose activity is weak, often dust-dominated, and therefore difficult to isolate with standard asteroid pipelines (Hsieh et al., 2014).

1. Survey setting and scientific role

PSComet was developed in the Pan-STARRS1 context as part of a survey that discovered two main-belt comets, P/2012 T1 (PANSTARRS) and P/2013 R3 (Catalina-PANSTARRS), while simultaneously accumulating a large statistical sample of main-belt detections. Its immediate role was operational: to identify detections that were inconsistent with point-like asteroid morphology and therefore warranted human review. Its broader role was scientific: to support quantitative statements about comet detection efficiency, MBC abundance, and the orbital conditions under which weak cometary activity is most detectable (Hsieh et al., 2014).

Within that framework, PSComet addressed a specific observational regime. Main-belt comets can exhibit extremely subtle activity, including slight PSF broadening, weak unresolved coma, or low-surface-brightness dust structures. Standard moving-object discovery pipelines are optimized for asteroids, not for low-level non-stellar structure. PSComet therefore functioned as an activity-screening layer imposed on top of the survey’s routine transient and tracklet processing.

A plausible implication is that PSComet should be understood less as a single classifier than as an operational interface between asteroid discovery infrastructure and comet-specific vetting. This interpretation is consistent with the way the system couples automated flagging to human screening and follow-up.

2. Detection formalism in IPP and MOPS

The core PSComet procedure was implemented in the Pan-STARRS1 Image Processing Pipeline (IPP) and Moving Object Processing System (MOPS). For each transient detection, IPP measured the second moments moments_xx and moments_yy, the fitted psf_major and psf_minor half-widths of the local stellar PSF, and a global data-quality metric psfquality on a 0–1 scale. From these quantities, the system defined the dimensionless extent parameter

$$\mathrm{psfextent}=\sqrt{\frac{\mathrm{moments\_xx}\times \mathrm{moments\_yy}}{\mathrm{psf\_major}\times \mathrm{psf\_minor}}}.$$

For a perfect point source with the same shape as the model PSF, $\mathrm{psfextent}\approx 1$. Trailing of a slowly moving object increases moments_xx and/or moments_yy slightly, pushing psfextent into the $\sim 1.5$–$2.7$ range, depending on S/N, filter, and date (Hsieh et al., 2014).

The flagging rule operated at the tracklet level rather than the single-detection level. Any tracklet, usually comprising at least two detections over about one hour, whose median psfquality exceeded $0.4$ and whose median psfextent exceeded a prescribed threshold set above the 99th percentile of asteroid psfextent values was sent for human review. This thresholding strategy made the system explicitly relative to the empirical asteroid population rather than to a purely analytic PSF model.

Operationally, this design reflects a compromise between completeness and false-positive control. Because moving objects are frequently trailed, and because trailing alone broadens the PSF, PSComet does not equate non-stellar morphology with activity. Instead, it identifies statistically anomalous broadening relative to the survey’s own asteroid baseline.

3. Human screening, false positives, and measured efficiency

PSComet produced between 200 and 2000 candidates per night, most of which were artifacts, including cosmic rays, diffraction spikes, bad pixel clusters, and field-star residuals. A trained watcher rapidly rejected obvious false positives, then stacked the remaining candidates for visual confirmation and decided whether to trigger follow-up astrometry and imaging on telescopes of at least 2 m class (Hsieh et al., 2014).

This human-in-the-loop architecture was not incidental. It was required because many of the morphologies of interest were close to the threshold of detectability, while many of the failure modes of wide-field imaging pipelines can mimic weak coma or dust. PSComet therefore relied on a two-stage decision process: automated candidate generation followed by expert triage.

The same survey interval enabled an empirical estimate of discovery efficiency. Pan-STARRS1 discovered 28 new comets during 2012 May–2013 November. Over that same interval, it also made observations that could have discovered 12 additional comets that were instead discovered by other surveys. Under the assumption that every opportunity was either a success or a miss,

$$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$

so that

$$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$

The paper explicitly identifies this 70% as an upper limit on the true detection efficiency, because some weak comets may never have been recognized by any survey (Hsieh et al., 2014). This distinction is important: the quoted efficiency constrains observed recognition, not the intrinsic sensitivity of the instrument to all possible low-activity bodies.

4. Complementary screening methods beyond radial PSF width

The Pan-STARRS1 analysis proposed six complementary methods to extend PSComet beyond the basic radial-PSF test. Each was framed as a way to flag detections or tracklets for the same downstream human screener.

Method	Core observable	Example flag criterion
Linear-PSF analysis	$$\mathrm{extent\_lin}=\sigma_{\mathrm{obs}}/\sigma_{\star}$$	$\mathrm{extent\_lin}\sim 1.2$–$1.3$
Extended-source photometry	$\mathrm{psfextent}\approx 1$0	$\mathrm{psfextent}\approx 1$1 mag
Photometric excess	$\mathrm{psfextent}\approx 1$2	$\mathrm{psfextent}\approx 1$3 mag

The first extension, linear-PSF analysis, targeted fast movers with apparent motion $\mathrm{psfextent}\approx 1$4, for which circular PSF tests lose sensitivity. The method rotates each detection to align the apparent motion vector horizontally, collapses the two-dimensional stamp onto its vertical axis, and compares the one-dimensional Gaussian width $\mathrm{psfextent}\approx 1$5 to the stellar width $\mathrm{psfextent}\approx 1$6. The resulting ratio, $\mathrm{psfextent}\approx 1$7, is flagged when it exceeds about $\mathrm{psfextent}\approx 1$8–$\mathrm{psfextent}\approx 1$9, depending on S/N (Hsieh et al., 2014).

The second, trailed-PSF modeling, treated asteroid PSFs as the convolution of the stellar PSF with a line segment proportional to the sky-plane velocity. A family of convolved PSFs is generated over a trial velocity range, compared to the observed stamp, and evaluated with $\sim 1.5$0. If the best-fit trailed PSF still fits significantly worse than a pure trailed model by more than a threshold $\sim 1.5$1, the source is flagged as a possible coma or dust-broadened detection. This method is specifically designed to separate activity from broadening induced solely by motion.

The third, azimuthal activity searches, or “Slice-&-Sum,” targeted thin dust tails that contain only a few percent of the total flux and can therefore evade PSF-based tests. For each stacked stamp, an annulus with inner radius about 3 PSF and outer radius about 6 PSF is subdivided into $\sim 1.5$2, with the example of 36 slices of $\sim 1.5$3. The slice fluxes are compared to the median slice flux and its dispersion, and a tail direction is flagged if the maximum slice flux exceeds the median by $\sim 1.5$4 (Hsieh et al., 2014).

The fourth, extended-source photometry comparisons, used the difference between Model_mag and PSF_mag. Stars and asteroids satisfy $\sim 1.5$5 mag, whereas comets are expected near $\sim 1.5$6–$\sim 1.5$7 mag. The system therefore flags detections with $\sim 1.5$8 mag.

The fifth, photometric excess searches, compared the observed magnitude to a predicted magnitude derived from the orbit and phase function, using an archive of well-measured $\sim 1.5$9 and $2.7$0 values. The quantity $2.7$1 is used to detect unresolved coma; objects brighter than predicted by more than 0.2 mag, i.e. $2.7$2 mag, are flagged (Hsieh et al., 2014).

The sixth, crowd-sourcing, was proposed as a web-based review layer in which citizen scientists label candidate stamps as “tail,” “coma,” “stellar,” or “artifact,” with professional review triggered once a consensus threshold is reached. The paper’s rationale is that unusual morphologies such as multiple tails or large diffuse comae can defeat automated classifiers while remaining recognizable to human inspectors.

Taken together, these proposals show that PSComet was not conceptually limited to a single scalar extent statistic. It was presented as a modular search strategy that could integrate radial, linear, azimuthal, photometric, model-based, and human-recognition cues.

5. Population inference and the physical context of MBC activity

The survey analysis used PSComet-enabled detections to infer the abundance of active objects in the outer main belt. The sample consisted of $2.7$3 unique outer-belt asteroids with $2.7$4–$2.7$5 AU and $2.7$6, observed at true anomalies $2.7$7, where activity should be detectable. Two MBCs were found, so $2.7$8. Assuming completeness $2.7$9 and a log-uniform prior

$0.4$0

the Poisson-likelihood approximation for large $0.4$1 is

$0.4$2

which yields the posterior

$0.4$3

From this posterior, the median fraction is $0.4$4 per $0.4$5 outer main-belt asteroids, and the 95% upper limit is $0.4$6 per $0.4$7 (Hsieh et al., 2014).

The same analysis reported an excess of high eccentricities, specifically $0.4$8, among known MBCs relative to the background asteroid population. The proposed physical explanation is eccentricity-driven sublimation modulation. In the grey-body energy balance without sublimation,

$0.4$9

with $$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$0, $$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$1, $$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$2, and $$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$3. With sublimation cooling, the balance becomes

$$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$4

where $$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$5, and the mass flux $$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$6 is determined by the vapor pressure $$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$7 (Hsieh et al., 2014).

For a canonical MBC with $$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$8 AU and $$N_{\mathrm{success}}=28,\qquad N_{\mathrm{missed}}=12,\qquad N_{\mathrm{total}}=40,$$9, corresponding to perihelion $$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$0 AU and aphelion $$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$1 AU, the isothermal solution gives

• $$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$2,
• $$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$3,

a variation by a factor of about $$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$4 over the orbit.

The paper further models mantle growth and enduring weak activity. Initially, the grain-size barrier gives a critical radius $$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$5–$$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$6 m. At $$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$7 AU, the instantaneous mantle-growth rate is about $$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$8 when the mantle is thin, so a diurnal-insulating layer of about 10 mm forms in about 0.1–1 yr. After a few orbits, the rubble mantle becomes much thinner than 1 mm yet still throttles activity to $$\epsilon_{\mathrm{obs}}=\frac{N_{\mathrm{success}}}{N_{\mathrm{total}}}=\frac{28}{40}=0.70.$$9–$$\mathrm{extent\_lin}=\sigma_{\mathrm{obs}}/\sigma_{\star}$$0 for decades (Hsieh et al., 2014).

This suggests that the objects sought by PSComet are not simply any active main-belt bodies, but preferentially those observed near orbital phases where sublimation is strongly amplified and where residual dust production remains detectable despite mantling.

6. Transfer to future surveys

The Pan-STARRS1 analysis explicitly considered how PSComet-style searches could be transferred to future facilities. The sensitivity scaling is given in terms of telescope diameter: a telescope of diameter $$\mathrm{extent\_lin}=\sigma_{\mathrm{obs}}/\sigma_{\star}$$1 collects photons proportional to $$\mathrm{extent\_lin}=\sigma_{\mathrm{obs}}/\sigma_{\star}$$2, so the coma-related signal-to-noise scales as $$\mathrm{extent\_lin}=\sigma_{\mathrm{obs}}/\sigma_{\star}$$3. Comparing LSST with $$\mathrm{extent\_lin}=\sigma_{\mathrm{obs}}/\sigma_{\star}$$4 m to PS1 with $$\mathrm{extent\_lin}=\sigma_{\mathrm{obs}}/\sigma_{\star}$$5 m gives an improvement of at least $$\mathrm{extent\_lin}=\sigma_{\mathrm{obs}}/\sigma_{\star}$$6 in psfextent sensitivity, corresponding to detection of comae about 5 times fainter (Hsieh et al., 2014).

The transfer pathway is algorithmic as well as instrumental. The proposed implementation includes early forcing of radial-PSF, linear-PSF, and extended-photometry tests in the nightly reduction pipeline; insertion of a short line-kernel convolution step for trailed-PSF fits; straightforward parallelization of azimuthal-slice code over $$\mathrm{extent\_lin}=\sigma_{\mathrm{obs}}/\sigma_{\star}$$7; and a per-tracklet lookup against a PSF_mag–VSOCmag archive for photometric-excess detection. Crowd-sourcing is described as wrapping the “maybe” stamps into a Zooniverse-style interface, with decisions stored in a small database.

The operational consequence is a substantially larger follow-up burden. The paper notes that a 2–4 m facility with a wide-field imager and queue scheduling will be needed to confirm a few hundred faint comet candidates nightly. It also states that combining all six screening methods with a robust follow-up plan and larger future apertures could scale PSComet-style searches up by 10–100 times in discovery rate (Hsieh et al., 2014).

In that sense, PSComet occupies an intermediate position in survey methodology. It is neither a purely automatic classifier nor merely a visual inspection protocol. It is a survey architecture for weak-activity discovery, built around anomaly detection in moving-object morphology, empirically calibrated against asteroid statistics, and extensible to deeper datasets and higher-throughput follow-up regimes.