heliostack: A Novel Approach to Minor Planet Discovery

1. Problem Statement

The detection of faint solar system objects (SSOs), such as distant Kuiper Belt Objects (KBOs), is limited by the sensitivity of individual telescope exposures. Conventional "shift-and-stack" techniques, which align and co-add images along an object's trajectory to boost signal-to-noise ratio (S/N), have historically been restricted to short time baselines (a few hours).

• The Limitation: Over longer time spans (days to weeks), the apparent motion of SSOs becomes nonlinear in pixel space due to parallax and Earth's orbital motion.
• The Computational Challenge: Searching for objects with nonlinear trajectories requires a six-dimensional orbital parameter space. Historically, the computational cost of enumerating and testing all possible trajectories over long baselines was prohibitive, limiting systematic searches to dedicated surveys with high-cadence, short-duration imaging.
• The Goal: To develop an algorithm capable of performing exhaustive, nonlinear shift-and-stack searches over long time baselines (e.g., 15 days) using existing archival data, thereby pushing detection limits deeper into the faint magnitude regime.

2. Methodology: The heliostack Algorithm

The authors introduce heliostack, an algorithm inspired by the HelioLinC linking method, designed to synchronize pixels from multiple images onto a common barycentric canvas based on asserted orbital parameters.

Core Concept

Instead of tracking pre-detected "tracklets," heliostack treats every pixel in every image as a potential detection. By asserting a set of orbital parameters, the algorithm calculates where a pixel from a specific time and location would appear on a reference "barycentric sky canvas" at a common epoch. If the parameters are correct, the signal from the object across all images will stack constructively at a single location on the canvas.

Technical Implementation

1. Parameterization: The algorithm asserts four orbital parameters to define a trajectory:
   • Barycentric distance ($r$)
   • Line-of-sight velocity ($v_r$)
   • Tangential velocity ($v_\Omega$)
   • Orbital inclination ($i$) and a sign parameter ($\kappa$) to break degeneracy.
   • Note: The right ascension ($\alpha$) and declination ($\delta$) are derived directly from the pixel's position in the image.
2. Orbital Propagation: Using the two-body problem formalism (Stumpff functions), the algorithm propagates the state vector (position and velocity) from the observation time to a reference epoch ($t_{ref}$).
3. Canvas Synchronization:
   • A "canvas" (a grid of barycentric coordinates) is created.
   • For each asserted orbit, a perspective transformation matrix is calculated to map topocentric image coordinates to the barycentric canvas.
   • Images are resampled and accumulated onto the canvas. The algorithm maintains running sums of the signal ($\xi$) and variance ($\zeta$) to compute a significance map ($\nu = \xi / \sqrt{\zeta}$).
4. Optimization:
   • Discrete Search: A grid of initial conditions is searched to find candidate peaks ($\nu \geq 7.0$).
   • Continuous Tune-up: Candidates undergo a 6-parameter orbit fit using a downhill simplex optimizer to maximize the log-likelihood ($\nu^2/2$) on the raw difference images, refining the orbital elements and flux.

3. Key Contributions

• Nonlinear Shift-and-Stack: heliostack is the first systematic implementation of shift-and-stack over time baselines significantly longer than one day (specifically 15 days in this demonstration).
• Pixel-Level Processing: By treating pixels as detections rather than relying on pre-existing catalogs, the method avoids the "tracklet loss" problem inherent in sparse datasets.
• Computational Efficiency: The algorithm leverages GPU acceleration and efficient perspective transformations (rather than pixel-by-pixel iteration) to make the search of $\sim 10^{14}$ trajectories feasible in a matter of hours.
• Heterogeneous Data Fusion: The framework allows for the combination of data from different telescopes or irregular cadences by transforming them to a common barycentric reference frame.

4. Results: Application to HST Archival Data

The algorithm was tested on archival Hubble Space Telescope (HST) data from a 2003 survey (G. M. Bernstein et al. 2004) targeting Cold Classical Kuiper Belt Objects (CCKBOs).

• Dataset: 1,118 images taken over a 15-day span, totaling $\sim$38,000 seconds of integration time.
• Search Space: 1,148,817 initial conditions covering the CCKBO parameter space ($a \in [42.4, 47.4]$ AU, $e < 0.2$, $i < 5^\circ$).
• Performance:
  • Recovery: Successfully recovered two known sub-threshold objects (2003 BF91 and 2003 BH91) that were previously detected but with lower significance.
  • New Discoveries: Identified two new objects (designated 2003 ABCD and 2003 WXYZ) that were too faint to be detected in the original analysis. These are the first objects discovered in stacks spanning $>1$ day.
  • Efficiency: The detection efficiency curve ($m_{50}$) reached 29.21 $\pm$ 0.03 mag, approximately 0.2 magnitudes deeper than the original 2004 analysis (which had $m_{50} \approx 29.17$).
  • False Positive Rate: Out of 145 final candidates, 141 were implanted synthetic sources (used for testing), and 4 were real objects. The search required no machine learning or visual inspection to filter false positives, demonstrating the high purity of the method on high-quality space-based data.

5. Significance and Future Impact

• Unlocking Archival Data: heliostack enables the re-analysis of existing deep surveys (e.g., HST, DES, LSST) to find faint objects previously missed due to short integration times per epoch.
• Future Surveys: The method is critical for upcoming missions like the Nancy Grace Roman Space Telescope and JWST, where observing time is scarce and cadences may be irregular. It allows for deep searches without requiring dedicated, high-cadence observing blocks.
• Targeted Science: The algorithm can be applied to specific populations (e.g., Planet Nine, Neptune Trojans, New Horizons flyby targets) by constraining the search space, making the computational cost predictable.
• Scalability: While currently limited by the cubic growth of required stacks with time span, the method is poised to become the standard for "trackletless linking" in sparse, long-baseline datasets as computing power continues to grow.

In conclusion, heliostack represents a paradigm shift in minor planet discovery, moving from linear, short-baseline stacking to a robust, nonlinear, long-baseline framework that maximizes the scientific return of deep astronomical surveys.