How ARCKEEP works

What ARCKEEP does with a series of telescope frames of an asteroid: how it finds where the frames point, which moment each position belongs to, how it follows the asteroid from frame to frame, how it predicts and identifies a known asteroid, and what it writes for the Minor Planet Center.

The page is written to be read two ways. The main text assumes first-year engineering: Newton's laws, vectors and basic statistics. The boxed notes marked "for researchers" give the exact method, its assumptions and its limits, so a specialist can judge it without reading the code. To learn the buttons rather than the method, read how to use ARCKEEP.

ARCKEEP is a proof of concept built by first-year space technology students at LTU and coded with AI agents. Every stage carries a status word. working means the capability runs in the website today for the inputs named. tested means it exists and has passed a recorded check, but the website does not use it yet. planned means it does not exist yet.

1. Reading the frames
2. From an image to a sky position
3. Which moment a position belongs to
4. Finding the asteroid in each frame
5. Following one moving object
6. Predicting a known asteroid
7. Which asteroid is this?
8. Reporting: the ADES file
9. How sure is a position? (summary)
10. Where your data goes
11. What ARCKEEP does not claim
12. Glossary
13. Credits and sources

1 Reading the frames working

A telescope frame arrives as a FITS file: a grid of pixel brightness values plus a header of text cards that say when, where and how it was taken. ARCKEEP reads the file in your browser. From the header it takes the observation time (DATE-OBS), the exposure length, the observing site's MPC code, the filter, and the telescope's rough pointing and pixel scale.

It then estimates the sky background across the frame and finds up to 200 sources: the stars and anything else brighter than the background. For each source it measures a centroid (the centre of the blob), its brightness and its shape.

Frames from a telescope that tracks the asteroid show the stars as short parallel trails. ARCKEEP recognises that, measures each trail's midpoint, and also runs a line filter that finds faint trails too weak to extract. Those faint ones are marked as leads for inspection only; they never enter the solve.

2 From an image to a sky position working

Science needs directions on the sky, given as right ascension (RA) and declination (Dec): the sky's longitude and latitude. ARCKEEP matches the pattern of stars in the frame against the Gaia star catalogue and fits the mapping between pixel coordinates and sky coordinates. That mapping is the plate solution, also called the WCS. Once it exists, any pixel, including the asteroid's, converts to RA and Dec.

The reference stars come from ESA's Gaia mission, the most precise star map there is. ARCKEEP built its own index of every Gaia star down to magnitude 17, 157,731,259 stars, and keeps it on Cloudflare storage. For each frame the browser asks only for a small circle of sky around the header's pointing, and only as deep as the frame needs: about twice as many stars as the frame shows. Fainter stars than the frame records would only slow the search and add false matches.

The first attempt uses only the catalogue stars that can lie on the frame. If it fails, ARCKEEP tries once more with the whole circle at full depth. Solving runs in background workers, so the page stays usable while a hard frame is searched.

3 Which moment a position belongs to working

An asteroid moves during the exposure, so a position is only meaningful together with the moment it belongs to: normally the middle of the exposure. FITS headers record a time, but do not say whether it marks the start or the middle. Guessing wrong puts the position half an exposure late, which for a fast asteroid is several arcseconds along its path.

ARCKEEP reads DATE-OBS as UTC and, by default, adds half the exposure to get the midpoint. A site moves to the mid-exposure list only when ARCKEEP's positions of known asteroids, compared with their MPC orbits, show that its DATE-OBS already marks the middle. Tautenburg (033) is the only such site so far. The Info view says which rule was applied to each frame, and says so when it is the unverified default.

4 Finding the asteroid in each frame working

In most frames the asteroid is the one source with a different shape. If the telescope followed the stars, the stars are round and a moving asteroid makes a streak. If the telescope followed the asteroid, the stars are streaks and the asteroid is round. ARCKEEP compares every source with the frame's typical shape and marks a target candidate only when exactly one source has the opposite shape. With none, or several, it leaves the choice to the track (section 5) or to you.

A long faint trail that the source finder breaks into pieces is still recognised as one line, with its ends and midpoint measured on the original pixels. A candidate is a candidate: shape alone never proves which object it is.

A whole series of frames can find what one frame cannot. Before solving, ARCKEEP can line the frames up on their stars and subtract what stays still: stars, galaxies and the sky's glow. What is left is sorted by how it moves. A spot that stays on the same pixel of the camera is a hot pixel. A long streak in one frame is a satellite or aircraft. An object that moves in a straight line at a steady speed across many frames is the candidate, and it becomes the target in each frame where it is detected. Its position is always measured on that frame's own pixels, never on the combined image.

5 Following one moving object working

An asteroid moves against the stars, so each frame in a series shows it in a slightly different place. ARCKEEP orders the frames by the middle of each exposure, converts each candidate to a sky position, and fits a straight line in time through those positions, separately in the two sky directions. Over minutes to an hour, an asteroid's path is very nearly a straight line at constant speed.

Each point's distance from the line is its O-C: observed minus computed. A point far from the line, or a frame where several candidates fit, is flagged. You must confirm or fix every flagged frame before the sequence can be confirmed. Plot and Verify shows the track as an animation, as motion on the sky, as position against time and as residuals, with a table of every frame.

The straight line fails for objects passing close to Earth. Their path curves within the hour, partly because the telescope itself is carried around by Earth's rotation. With five or more frames ARCKEEP also fits a curve and warns when the bend is significant. The lasting check is against the predicted path (sections 6 and 7).

6 Predicting where a known asteroid is working

ARCKEEP computes, in your browser, where a catalogued asteroid should appear from your site at each frame's time. It is used three ways: to check a named object against its orbit, to search for which known object is in the frames, and to draw a candidate's path on a frame (section 7).

6.1 An orbit is six numbers at a moment

To know where an asteroid will be, you need where it is and how it is moving at one instant: a position vector (x, y, z) and a velocity vector (vx, vy, vz) relative to the Sun. With those six numbers and the forces acting on it, Newton's second law gives its position at any other time. The Minor Planet Center (MPC) publishes these six numbers, with their uncertainties, for every catalogued asteroid. Together they are the state vector.

6.2 The Sun alone gives an exact answer

If the Sun were the only body pulling, the asteroid would follow a fixed ellipse forever. That is the two-body problem, and it has an exact solution: Kepler's equation says where on the ellipse the asteroid is at any time. No step-by-step simulation is needed, and no error builds up.

6.3 The planets add nudges

The planets pull too: mostly Jupiter, and Earth for objects passing close. Their pull slowly bends and reshapes the ellipse. The method handles this in short time steps:

1. Coast along the exact Sun-only ellipse for a short step.
2. Give the velocity a small kick equal to the planets' pull over that step.
3. Repeat.

It is like a car on cruise control along a curved road, with small steering corrections now and then. Close to Earth, where Earth's pull changes quickly, the steps are made much shorter. The planets' positions come from JPL's DE440 ephemeris, stored with the website for 2010 to 2040.

6.4 Why not just integrate x, y and z?

You could integrate the six numbers directly with a standard numerical method. The worry is that tiny rounding errors accumulate step after step, and the orbit slowly drifts.

Under the Sun alone, the size, shape and tilt of the orbit never change; only the asteroid's position along it does. Orbital elements (semi-major axis, eccentricity, inclination and so on) describe exactly those fixed properties, so the coasting step changes one number and holds the rest fixed. Nothing can drift during the coast.

The planets, however, really do change the orbit. So the method coasts exactly on the current orbit and lets the planetary kicks move it to the next one. It does not force the asteroid back onto the orbit it started on, because that would delete real physics.

6.5 From a position in space to a position on the sky

The prediction gives the asteroid's position relative to the Sun. The telescope sees a direction. To convert:

1. Work out where the telescope is: Earth's position, plus the site's position on the rotating, slowly wobbling Earth.
2. Subtract the telescope's position from the asteroid's position.
3. Correct for light travel time. You see the asteroid where it was when the light left it, seconds to minutes earlier for a nearby asteroid.
4. Convert the resulting direction to RA and Dec.

7 Which known asteroid is this? working

A confirmed track says where an object was, how fast it moved and in which direction, from a known site at known times. ARCKEEP uses that three ways, all in Plot and Verify or the frame's Info view.

7.1 Automatic identification

When Plot and Verify opens, ARCKEEP searches every near-Earth asteroid the MPC lists, about 42,800 of them, in two steps:

1. Sieve. A fast Sun-only calculation for every object, keeping only those that could be near the frames. In the recorded test cases, 58 to 115 of 42,465 orbits survived.
2. Gate. The full calculation of section 6 for each survivor, at every frame's time, compared with the measured positions: both where the object is and how it moves.

Comparing motion as well as position matters. Two asteroids can sit at the same spot by chance, but rarely with the same motion too. It works like matching a speed-camera photo to a reported car: location, heading and speed must all agree. If exactly one object passes, ARCKEEP checks it against its full-precision orbit (7.2). If several pass, it lists them and lets you choose; it never picks between them. If none passes, it says what was not searched.

The search has limits. It covers near-Earth asteroids only, and only frames within 200 days of the date the MPC's orbit file refers to. Older archive frames, such as the 2019 examples in the guide, are reported as out of reach, and MPChecker is the way to find candidates for them.

7.2 Check against a named object's orbit

Type a designation and ARCKEEP fetches that object's orbit from the MPC, predicts its position at every frame time, and reports how far the measured positions sit from it: along the direction of motion and across it. It then gives one of three answers: consistent; the time tags look off, when every position is shifted along the motion by the same amount; or the orbit is too uncertain to judge. Positions that do not match block the sequence.

The along-track shift is the clock check. A shift that is the same in every frame, divided by the asteroid's speed on the sky, is a time error. This is how ARCKEEP found Tautenburg's time convention (section 3).

7.3 Compare paths on one frame

In a frame's Info view, paste up to 12 designations (from MPChecker, for example) and ARCKEEP draws each one's predicted path across the frame, minute by minute around the exposure, and says which paths cross it. Position alone does not identify a source; the path tells you where to look.

8 Reporting: the ADES file working

The MPC collects asteroid positions from observers worldwide in a format called ADES. Once a sequence is confirmed, ARCKEEP writes one ADES file for it, recording for each frame the time, position, observing site, catalogue, exposure and plate-fit evidence, plus the submitter, observers, measurers and telescope, filled from the FITS header where it names them. By default it writes five evenly spaced positions, because the MPC asks for three to five per object per night. It never sends the file anywhere; you review it, copy or download it, and submit it yourself.

ADES has optional fields for the uncertainty of each position. ARCKEEP leaves them out on purpose. It has no calibrated way to fill them yet, and an invented uncertainty is worse than none, because the MPC would weight the measurement by it. Section 9 describes how that changes.

8a Starting from an ADES file tested

An ADES file already holds what the identification compares: time, RA, Dec and site for each observation. ARCKEEP has a reader for ADES files and can compare every observation with an orbit. The website cannot load an ADES file yet.

9 Uncertainty: how sure is a position? planned

ARCKEEP reports no position uncertainty today (section 8). Two different uncertainties matter, and both are needed:

• Measurement uncertainty: how well we measured where the asteroid appeared. It comes from the plate solution, the centroid of the blob and the timing, combined in quadrature, and it is an ellipse rather than a circle.
• Prediction uncertainty: how well we know where the asteroid should be. Tiny for a well-observed asteroid, huge for one seen on a single night, because one night does not tell you the distance.

For a well-known asteroid the prediction is far more precise than any measurement, so measuring known asteroids calibrates ARCKEEP's own error bars. The first pieces exist: the track's per-axis scatter and rate error (section 5) and the known-object check (section 7.2). Only after calibration will ARCKEEP report uncertainties.

Read the full explanation on the uncertainty page.

10 Where your data goes

Your FITS files never leave your browser. They are read, solved and measured on your own computer, and remembered in your browser's own storage so that leaving the page loses nothing. Removing a frame or closing a workspace deletes it there too.

ARCKEEP's server receives only what it needs to fetch reference data:

The status page shows whether each outside service is answering. When one is down, the engine's terminal names it and what it affects, so a failed solve is not mistaken for a bad frame.

11 What ARCKEEP does not claim

• A position that fits a track, a shape or a predicted path is consistent with an object. It is never proof of identity.
• Plate-fit RMS is not target accuracy.
• A schema-valid ADES file is not an accepted submission. The MPC decides acceptance.
• Identification covers near-Earth asteroids only, and only frames within 200 days of the MPC orbit file's epoch.
• ARCKEEP does not link unknown detections into new objects or search for lost ones.
• ARCKEEP is a proof of concept. Check its output before relying on it.

12 Glossary

13 Credits and sources

ARCKEEP stands on published methods, open data and other people's software. This section credits what each stage uses. The full register, with exact code locations and the reason each source is in or out of the product, is the repository's docs/REFERENCES.md.

Used in ARCKEEP today

Behind the planned stages

JPL Horizons is used only by hand, to cross-check predictions during validation, as JPL's API policy requires.