🔭 astrophysics

Independent Recovery of Vanishing Sources on POSS-I Photographic Plates Using Automated Source Detection and Cross-Epoch Matching

This paper presents an automated pipeline that successfully identifies a catalog of 2.85 million candidate vanished sources on POSS-I plates with strong replication against existing catalogs, though subsequent temporal analysis fails to find a statistically significant association with transient events.

Original authors: Zachary Hayes

Published 2026-04-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Zachary Hayes

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a detective trying to solve a mystery that happened 70 years ago. The crime scene is the night sky, and the "suspects" are stars that seem to have vanished into thin air.

This paper is the report from a new, independent detective (Zachary Hayes) who built a robot to re-investigate these missing stars using old, digitized photographs of the sky.

Here is the story of the investigation, broken down simply:

1. The Mystery: "Vanishing" Stars

Back in the 1950s, astronomers took photos of the sky on glass plates. Recently, other scientists (the VASCO team) found thousands of spots on these old photos that look like stars but don't exist in modern telescopes.

The Big Question: Did these stars actually disappear? Or are they just glitches in the old photos?
The Wild Theory: Some researchers suggested these "vanishing stars" might be linked to nuclear bomb tests from the 1950s. The idea was that the flashes from the tests might have left marks on the film that look like stars, but only for a split second.

2. The New Detective's Tool: A "Star-Sniffer" Robot

The author didn't trust the old methods, so he built a completely new, automated computer program (a pipeline) to find these missing stars from scratch.

How it works: Imagine the robot is looking at a stack of three photo albums:
1. Album A (Red): The old 1950s photo.
2. Album B (Blue): The old 1950s photo taken at the same time but with a different filter.
3. Album C (Modern): A high-definition photo from today.
The Rule: If the robot sees a "star" in Album A, but doesn't see it in Album B or Album C, it flags it as a "Vanishing Candidate."
The Filter: The robot is very picky. It only looks for perfect, round dots (like real stars) and ignores smudges, scratches, or weird shapes (like galaxies or dust).

3. The Test Drive: Did the Robot Work?

Before scanning the whole sky, the author tested the robot on two famous "crime scenes" where the missing stars were already known.

The Result: The robot found 8 out of 9 missing stars in one case and all 3 in another.
The "False Alarm" Check: When the robot looked at random, boring parts of the sky with no missing stars, it rarely made mistakes (only about 1 fake alarm per 10 fields).
The Verdict: The robot is sharp. It can independently find these mysterious objects without needing the original list of suspects.

4. The Big Sweep: Finding Millions of Suspects

The robot then scanned the entire sky covered by the old 1950s photos.

The Findings: It started with over 6 million "suspicious" spots. After cleaning up the list (removing duplicates and obvious errors), it ended up with 2.85 million candidates.
Cross-Checking: The author compared his list with the original VASCO list. They matched up about 64% of the time. This is a huge success! It proves that the "vanishing stars" are real phenomena in the data, not just a fluke of one specific computer program.
The Modern Check: The author checked these candidates against modern, super-clear telescopes (Pan-STARRS). Almost none of them showed up today. This confirms they are truly "missing" from the modern sky, though it doesn't prove why they are missing.

5. The Nuclear Bomb Theory: Did the Tests Cause It?

This is the most controversial part. The author checked if the "vanishing stars" appeared more often right after nuclear bomb tests.

The "Calendar" Look: If you just look at the calendar, there seems to be a slight bump in vanishing stars the day after a test. It looks like a pattern.
The "Real" Look (The Math): But when the author did the deep math (accounting for how many photos were actually taken each night), the pattern disappeared.
The Analogy: Imagine you are counting how many people wear red hats at a park. You notice more red hats on Tuesdays. But then you realize: The park is only open on Tuesdays. Of course, you see more red hats on Tuesdays!
- In this study, the "park" (the sky survey) was only open on certain nights. The "red hats" (vanishing stars) were found on every single night the survey happened.
- Because the robot found a "vanishing star" on every single night it looked, the math couldn't tell if the bombs caused them or if the robot was just finding them because it was looking.

The Final Conclusion

Good News: We have a new, reliable way to find these "vanishing stars." The robot confirms that thousands of these objects exist in the old photos and are missing from modern ones.
Bad News (for the theory): The evidence linking these stars to nuclear bomb tests is weak and inconclusive. The apparent connection was likely just a side effect of when the photos were taken, not the bombs themselves.
The Future: To solve the nuclear mystery, we need to clean up the data even more (using the robot's stricter settings) and look at the numbers very carefully. For now, the "vanishing stars" remain a fascinating mystery, but they probably aren't caused by atomic bombs.

In short: The robot is great at finding the missing stars, but it didn't find the smoking gun for the nuclear theory. The mystery of the vanishing stars continues!

1. Problem Statement

The VASCO (Vanishing and Appearing Sources during a Century of Observations) project has identified thousands of candidate transient sources by comparing historical Palomar Observatory Sky Survey (POSS-I) photographic plates with modern digital archives. Notable findings include simultaneous appearing/vanishing sources in 1950 and 1952, and a catalog of 5,399 vanishing candidates.

The Debate: The nature of these transients is contested. Some suggest they are emulsion flaws, while others argue they are sub-second optical flashes.
The Controversy: A recent study (Bruehl & Villarroel, 2025) reported a statistical association between the occurrence of these transients and above-ground nuclear weapons tests (1949–1957), suggesting potential environmental contamination rather than astrophysical phenomena.
The Need: Independent replication using a different detection pipeline is essential to validate the robustness of these catalogs and the validity of the temporal correlation with nuclear tests.

2. Methodology

The author developed a fully automated, independent pipeline that detects sources without referencing existing transient catalogs. The workflow consists of the following stages:

Data Sources: Digitized POSS-I Red and Blue plates (1949–1957) and POSS-II Red (1985–2000) from the STScI Digitized Sky Survey (DSS), plus Pan-STARRS DR1 for verification.
Source Detection & Filtering:
- Sources are detected via sigma-clipped thresholding and connected-component labeling.
- PSF Morphology Filter: Retains only point-like sources based on Full Width at Half Maximum (FWHM) and ellipticity relative to the plate median. This rejects extended objects, plate artifacts, and noise.
- Calibration: Benchmarked on 20 cases, the optimal parameters were set to $8\sigma$ threshold, FWHM $0.7–1.5\times$ median, and ellipticity $<0.3$ .
Astrometric Refinement:
- Global WCS headers from DSS often have $1–3$ arcsec residuals.
- The pipeline computes a local 6-parameter affine transformation using the 20–50 brightest isolated sources between epochs, reducing residuals to a median of $0.2–0.5$ arcsec.
Transient Identification:
- A source is flagged as a candidate if detected on POSS-I Red but absent on both POSS-I Blue and POSS-II Red within an $8$ arcsec radius (after correction).
- This dual-epoch absence requirement minimizes false positives from color-dependent detection limits.
Pan-STARRS Verification:
- Candidates are cross-matched against Pan-STARRS DR1. Sources with a counterpart within $3$ arcsec are rejected as persistent objects that failed matching due to morphological differences (plate vs. CCD).
Synthetic Injection Testing:
- Artificial stars and artifacts were injected to test completeness. The pipeline recovered $79.2\%$ of injected stars and rejected $96.7\%$ of artifacts, confirming utility for validation but noting it is not exhaustive for population-level claims.

3. Key Results

A. Benchmark Validation

The pipeline was tested against known benchmark fields:

April 1950 Field: Recovered 8/9 known transients.
July 1952 Field: Recovered 3/3 known transients.
Control Fields: Random non-crowded fields showed a false positive (FP) rate of $\sim0.2$ per $10$ arcmin. Crowded Galactic plane fields showed higher FP rates (up to 22 per field) due to source confusion.

B. Cross-Match with Solano et al. (2022) Catalog

The pipeline was applied to a full-sky sweep ($122,991$ patches, $932$ plates), generating a preliminary catalog of $6.6$ million candidates, which was filtered down to 2.85 million candidate vanished sources.

Recovery Rate: Cross-matching against the public Solano et al. (2022) catalog ($5,399$ sources) yielded 3,450 matches (63.9%).
Positional Accuracy: The median angular separation between matches was 0.94 arcsec, with $98.7\%$ of matches within $3$ arcsec.
Analysis of Unrecovered Sources:
- $54\%$ of unrecovered Solano sources were detected in the raw pipeline data but rejected by the PSF morphology filter (suggesting methodological differences in source shape criteria).
- $46\%$ showed no detection above the $5\sigma$ threshold (suggesting they are below the detection limit or catalog artifacts).
- Crucially: None of the unrecovered sources within the survey footprint had a Pan-STARRS counterpart within $3$ arcsec, supporting the "archive-absence" classification.

C. Temporal Correlation with Nuclear Tests

The study tested the hypothesis that transient counts correlate with nuclear test dates (1949–1957) using two statistical approaches:

Calendar-Day Binary Analysis:
- Compared the presence of at least one transient on days relative to nuclear tests.
- Found a Relative Risk (RR) of 1.35 for the day immediately following a test ( $[+1]$ window) compared to baseline.
- Significance: The result was not statistically significant (Two-sided Fisher $p=0.17$ ; 95% CI $[0.91, 2.00]$ ). The authors note this metric is sensitive to the survey schedule, as every observation night in the study window contained at least one candidate.
Negative Binomial GLM (Count Model):
- Modeled the number of candidates per night, using nightly patch coverage as an exposure term to account for varying sky coverage.
- Result: The model was null for all tested windows. For the symmetric $\pm1$ day window, the Incidence Rate Ratio (IRR) was 1.03 ( $p=0.71$ ).

4. Key Contributions

Independent Pipeline: Established a fully automated, catalog-agnostic pipeline for detecting vanished sources on POSS-I plates.
Validation of VASCO Catalog: Confirmed that $\sim64\%$ of the Solano et al. (2022) catalog can be independently recovered with sub-arcsecond precision, validating the existence of a large population of "vanished" sources.
Methodological Insight: Identified that a significant portion of unrecovered catalog entries are rejected due to PSF morphology differences, highlighting the sensitivity of transient detection to source shape criteria.
Temporal Analysis: Provided a rigorous statistical test of the nuclear test correlation using a new, larger dataset.

5. Significance and Conclusions

Catalog Replication: The study strongly replicates the catalog-level findings of the VASCO project, confirming that a substantial number of sources on POSS-I plates lack counterparts in modern archives (POSS-II and Pan-STARRS).
Temporal Association: The study fails to independently confirm the statistical association between transient rates and nuclear tests found by Bruehl & Villarroel (2025).
- The "post-test elevation" observed in binary day analysis is likely an artifact of the survey schedule (observation nights always having candidates) rather than a physical effect.
- The more robust count-based model (Negative Binomial GLM) shows no significant correlation.
Future Work: The authors conclude that the temporal question remains open but requires a cleaner full-sky reprocessing using the calibrated $8\sigma$ parameters (the current sweep used preliminary $5\sigma$ settings) and explicit control for observational confounds.

Summary: The paper successfully validates the existence of the vanishing source catalog through independent detection but casts significant doubt on the nuclear test hypothesis when analyzed with a more rigorous statistical model and a larger, independently generated dataset.