Little red dots — three sky maps

Little red dots are compact, red, high-redshift sources that JWST found in numbers nobody expected. These three maps answer a narrower question than what they are: where they are, and who has looked.

The first two maps are drawn from published catalog positions; the third from positions measured in the images. No third-party selection is re-derived.

The published population

Sky map of 1745 published JWST little red dot catalog entries on both celestial hemispheres, with zoomed panels on the six deep fields that contain 96 percent of them. — Click any figure for the full-resolution version. Nine catalogs, vendored so the figure cannot drift when a URL does.

1745 catalog entries across 34 distinct pointings, in about 0.005% of the sky. Six deep fields hold 96% of them. Every count on these figures is a catalog entry; the 1745 entries are 1124 distinct objects at a one-arcsecond match, because the same source is published by several papers. That concentration is the single most important fact about the sample: it is a census of six patches, not of the sky.

Without JWST

3368 catalog entries over 0.265% of the sky at redshifts 0.33 to 3.66 — about fifty-five times more sky than the JWST card covers, at roughly a tenth the redshift. They are kept on their own map deliberately. Dropped in beside the JWST population they would outnumber it two to one and quietly change what "every published little red dot" means.

The two populations do not share a single object. No published JWST little red dot lies within five arcseconds of one found without JWST; the closest approach anywhere on the sky is 86 arcseconds. Euclid's Fornax field does cover the same sky as GOODS-S — 335 Euclid objects against 132 JWST ones inside the same box — and even there the nearest pair is 635 arcseconds apart.

That is the sanity check passing rather than a surprising null. The two searches are cut to find different objects. Euclid's Q1 sample spans redshift 0.33 to 3.66, and 96% of the published JWST little red dots that carry a measured redshift sit above that ceiling. The JWST population is also faint — a median of 25.8 in F444W — which is well below what a wide survey of tens of square degrees reaches. Selections that do not overlap in redshift or in depth cannot return the same objects, so a shared object would have been the thing worth explaining.

Two of the three Euclid fields are hollow in the middle. Measuring against Q1's own published region files rather than against a circle, Deep Field North holds 7.9 candidates per square degree inside 1.6 degrees of its center and 58.9 outside it, and Fornax 7.7 against 78.5. South is comparatively flat, 48.9 against 74.3.

The deficit is in the catalog, not in the sky. The Euclid imaging of those centers is full: the Q1 merged catalog holds 476,600 detected sources per square degree inside North's void against 474,500 outside it, and 405,400 against 418,600 in Fornax — a flat parent population either side of a boundary the candidates respect absolutely. Ancillary ground-based photometry covers 99.6% or more of those sources on both sides. The region files rasterize to one connected area per field, center covered, no interior hole. Gaia's bright-star density is flat across the boundary to within 5%, and the Galactic dust column at those two centers is lower than at their edges. An independent red-source cut on the same catalog — H at least five times the VIS flux, detected at 5σ — is flat across the same boundary, matching inside to outside within 6%, and within 7% for red point-like sources.

Nothing is obscuring that sky and nothing is missing from it. Anything in front of it — dust, a galaxy, a disc seen edge-on — would take the half-million ordinary sources per square degree down together with the red ones, and none of them are gone. What fails in the inner 1.6 degrees of those two fields is the published selection, not the observation. Those centers are not empty sky, and that catalog's sky distribution should not be read as one until the selection is understood.

Found independently

Candidates from an independent search of 89 JWST fields, graded by how far each one has been vetted. One field is excluded outright, for the reason set out below the map.

Sky map of 1555 little red dot candidates found by an independent search pipeline across 89 JWST fields, colored by confidence tier. — One pipeline's own objects, colored by how far each one has been vetted.

An independent search over 89 JWST fields, mostly lensing clusters and pure parallels, most of which nobody has published little red dots in. Positions are measured from the images rather than taken from a catalog.

The colors are the point. 64 entries, 38 distinct objects, are independent recoveries of somebody's published little red dot, which is the part that can be checked against the literature. The rest are graded, and the grades mean what they say — a pick is a pick, and a watch is not a claim.

One field is excluded. JWST program 12496 targets Centaurus A, 3.8 megaparsecs away, where JWST resolves individual stars. Its reddened red giants are compact, red and bright — exactly the selection box. 2955 met it; all are excluded from this map and from the search.

One of them, close up

The maps are populations. This is one object. MoM-BH*-1 sits at redshift 7.7569 — 660 million years after the Big Bang — and neither a star nor an ordinary quasar accounts for its spectrum. The model put forward for it is a black hole wearing a star's atmosphere: dense, turbulent, dust-free gas around a supermassive black hole, radiating from a surface some 792 astronomical units across. That is an interpretation of the light, not a picture of the object. It is not a one-off: the same signature has since been found in hundreds of other objects.

Four minutes, narrated. Eight parts: the actual NIRCam frames, in which the object is absent in the bluest band and unmistakable in the red; the same light band by band; the spectrum, with the Balmer break set against the strongest break a real stellar population can make; the model cut open and then rebuilt shell by shell; the ten days a photon takes to scatter out of the 42 AU skin, against the 192 it would need across the whole star; three readings of the same black hole mass that differ by two orders of magnitude; whether it can grow fast enough in the time available; and what all of that would mean for little red dots as a class. Each frame names its source. Physical numbers are from Naidu et al. (2026), Nature (preprint); the photometry is measured here from the released frames; the cut-open view is a rendering of the model, not an image of the object, and its radius is logarithmic — the horizon, the last stable orbit and the 792 AU photosphere are each a computed number for this object, four decades apart.

If you would rather not watch anything, the same material is one still image (smaller copy) carrying six panels, tagged the same way.

What the model implies for the dots on the maps above

If the black-hole-star model holds for the class and not just for this object, four things about little red dots that have been hard to explain at once stop being separate puzzles. The V-shaped spectral energy distribution that defines them is a blue host galaxy in the ultraviolet plus a red central engine in the optical — with no dust needed to redden anything. Measured on this object, the ultraviolet bands blueward of the break give a slope of −1.25 ± 0.28, and carrying that law redward leaves the blue component accounting for about 7% of the light at 4.4 µm. Their X-ray weakness is Compton-thick gas swallowing the X-rays before they get out. Their missing hot dust is because the red is intrinsic, so there is no torus left to glow. And they are always unresolved because the emitting surface is only about 792 astronomical units across. On this reading a little red dot is not a new kind of galaxy; it is an ordinary galaxy with something extraordinary in the middle. And if the engine's share is a dial rather than a switch, the color cut that defines the class samples one end of it: a 412-source census over redshift 2 to 11 finds that extreme cuts isolate under a quarter of the population, with a continuous range of spectral shapes behind them. That is the model's claim, and the caveats are the model's too: dust is now implicated in about half the population, most of the Balmer absorbers are outflowing rather than static, and the spectrum that decides it is not public until August 2027.

The mass is where the model bites hardest, and the arithmetic here is model-independent. Read the broad line width as orbital motion and the black hole is heavy — 10^8.3 solar masses, feeding at a quarter of the Eddington limit. Read it as electron scattering and it is light — about 10⁶, feeding at roughly five times Eddington. Those are the paper's own Table ED 5 readings. Run each forward at standard Salpeter growth from a 10^6.3 seed to a billion-solar-mass quasar, 6.2 e-foldings, and the heavy readings need 1.2 and 1.9 billion years. The universe was 664 million years old at this redshift. The super-Eddington reading does it in 62 million years, with room to spare. So if the line width is scattering, these are not overmassive black holes at all; they are modest ones eating faster than the usual speed limit, and the fast-feeding reading is the one that fits the clock.

There is a further consequence, and it is worth stating with its limit attached. If some of the rest-optical light in these systems comes from the engine rather than from stars, then fitting them with ordinary galaxy templates counts black-hole light as starlight and the inferred stellar mass comes out too high. That is a real effect, but it is not the whole of the early massive-galaxy problem: a spectroscopic census of the 200 most massive galaxies at these redshifts removes little red dots and broad-line AGN first, and still finds a massive population. The black hole side is cleaner — there, lighter holes and faster feeding remove the difficulty rather than reducing it.

Nor is this only an early-Universe story. A photometric search for the same signature has now found 241 black-hole-star candidates spread from redshift 9.3 down to 1.7 — the class runs from the first few hundred million years all the way to cosmic noon, some four billion years later.

Sources

JWST: Akins+24, Kocevski+24, Kokorev+24, Leung+24, Rinaldi+24, de Graaff+25, Barro+25, Weibel+26 and Matthee+26. Without JWST: Bisigello et al. for the Euclid Collaboration (Q1, arXiv:2503.15323, A&A 711, A24, via VizieR J/A+A/711/A24) and Lin et al. 2026 (DESI DR1, Zenodo 10.5281/zenodo.20309303). Each is taken from the authors' own machine-readable table where one exists, and from the paper's source where it does not.

The search pipeline and the candidate tables behind the third map are not public. For code, please contact jordan@cerinamroth.com.