Extragalactic Astrophysics
Ultra-Diffuse Galaxy
As wide as the Milky Way but a thousand times fainter — and a handful of them seem to have shed their dark matter
An ultra-diffuse galaxy (UDG) is a low-surface-brightness galaxy whose effective radius rivals that of the Milky Way (R_eff > 1.5 kpc) but whose stellar luminosity is roughly a thousand times lower. The class was named by van Dokkum and collaborators in 2015 after the Dragonfly Telephoto Array found about a thousand of them lurking in the Coma cluster. A few — most notoriously NGC 1052-DF2 and DF4 — appear to contain almost no dark matter, a result without precedent in any other galaxy that has been measured.
- Discoveryvan Dokkum et al., 2015
- Effective radiusR_eff > 1.5 kpc
- Surface brightnessμ_g(0) > 24 mag/arcsec²
- Coma count≈ 1000 candidates
- Famous DM-freeNGC 1052-DF2, DF4
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
What makes a galaxy "ultra-diffuse"
The term is operational, not theoretical. A galaxy qualifies as ultra-diffuse if it satisfies two thresholds simultaneously: a large half-light radius and a faint central surface brightness. The canonical cuts, from the van Dokkum et al. (2015) discovery paper, are R_eff > 1.5 kpc (with some authors raising the floor to 2 or 2.5 kpc) and a central g-band surface brightness fainter than μ_g(0) ≈ 24 mag/arcsec². Put concretely, that is about a hundredth of the central surface brightness of the Milky Way's disk averaged over its central few kiloparsecs — and yet the half-light radius is the same order of magnitude. The result is a galaxy that, for its size, contains roughly a thousand times less light than the Milky Way.
The thresholds are chosen to carve out objects that are hard to detect in conventional surveys. Sloan Digital Sky Survey pipelines and similar wide-field tools are tuned for high-contrast extended sources; an object with the surface brightness of a UDG looks like sky noise and is routinely missed or removed as a flat-field artefact. That is why the class went un-named until 2015, despite individual examples having been catalogued for decades — Sandage and Binggeli identified low-surface-brightness "dwarf" candidates in Virgo in the 1980s, but without a unifying observational program they were treated as oddities, not as a population.
The Dragonfly Telephoto Array
The instrument that turned UDGs from oddities into a population is unusual. The Dragonfly Telephoto Array, designed by Roberto Abraham and Pieter van Dokkum, is not a single telescope: it is a wide-field array of off-the-shelf Canon 400 mm f/2.8 telephoto camera lenses with nano-fabricated anti-reflection coatings, mounted on a common frame so their fields stack coherently. Reflective optics — even of professional observatory class — scatter light off micro-surface roughness and produce extended halos around bright stars that swamp diffuse low-surface-brightness signals. The lens coatings used in Dragonfly were originally developed for the photographic market and reduce internal reflections by orders of magnitude, giving the instrument a vastly cleaner point-spread function in the wings than any traditional telescope.
The Coma cluster pointing in van Dokkum et al. (2015) revealed about 47 large, faint galaxies — extended on arcminute scales but invisible in archival Sloan data. A follow-up program by Koda et al. (2015) using the Subaru Telescope confirmed and extended the catalogue to roughly 850 candidates over the full Coma area, and the count has since grown past a thousand. The same techniques applied to Virgo (the closer rich cluster) have found similarly large UDG populations, and field surveys with HST, MATLAS, and Euclid commissioning data are now extending the census to the lower-density universe.
A worked comparison: Milky Way versus a typical UDG
It is worth grounding the numbers in something familiar. Below, the Milky Way is compared to a representative UDG in the Coma cluster (Dragonfly-44 is used as a high-mass example; most UDGs are fainter still).
| Property | Milky Way | UDG (Dragonfly-44) | Ratio |
|---|---|---|---|
| Effective radius R_eff | ~ 4 kpc (disk scale) | ~ 4.6 kpc | ≈ 1× |
| Stellar mass M* | ~ 6 × 10¹⁰ M☉ | ~ 3 × 10⁸ M☉ | ~ 200× less |
| Total luminosity | ~ 3 × 10¹⁰ L☉ | ~ 3 × 10⁷ L☉ | ~ 1000× less |
| Central surface brightness μ₀ | ~ 21 mag/arcsec² (B) | ~ 25 mag/arcsec² (g) | ~ 40× fainter |
| Inferred halo mass | ~ 10¹² M☉ | ~ 10¹¹ M☉ (claimed) | ~ 10× less |
| Globular cluster count | ~ 150 | ~ 70 – 100 (disputed) | ~ 1× |
Two things stand out. First, the half-light radii are comparable — these objects are not small galaxies, they are full-size galaxies. Second, even the dwarf-galaxy mass (a few × 10⁸ M☉) and the globular cluster count is anomalously high for the stellar mass: a Milky-Way GC count in a luminosity-class-VI dwarf. That single mismatch motivated some early authors to argue that UDGs sit in unusually massive dark-matter halos, which is what made the later "no dark matter" result so jarring.
NGC 1052-DF2: the galaxy with no dark matter
In 2018 the van Dokkum group published a Nature paper on NGC 1052-DF2, a UDG in the NGC 1052 group, that used the radial velocities of ten luminous globular clusters as kinematic tracers. The line-of-sight velocity dispersion was σ_los ≈ 8.4 km/s, much smaller than the σ ≈ 30 km/s expected from a normal dwarf-galaxy dark-matter halo at that stellar mass. Translating that dispersion into a dynamical mass with the standard tracer-mass estimator gave M_dyn(< 7.6 kpc) ≈ 3.4 × 10⁸ M☉, essentially equal to the stellar mass of ≈ 2 × 10⁸ M☉ inferred from photometry. The ratio of dynamical to stellar mass is therefore close to unity — there is no room for the dominant dark-matter component that every other galaxy at this mass contains by a factor of ten or more.
A second galaxy in the same group, NGC 1052-DF4, was studied with the same method and produced a consistent result: σ_los ≈ 5.8 km/s, again indistinguishable from a stars-only system. The two galaxies are within a few hundred kiloparsecs of each other in the NGC 1052 group, hinting that whatever process produced them acted on both.
The result triggered an intense follow-up literature. Critics challenged the distance estimate (a closer DF2 would have proportionally more dark matter); the assumed velocity-anisotropy parameter β; the number of GCs used as tracers; and the GC luminosity function itself (its peak in DF2 is offset from the canonical universal value, which is either an intrinsic anomaly or an indication that the distance is wrong). Subsequent measurements with the Hubble Space Telescope tip-of-the-red-giant-branch method, with VLT spectroscopy of the stars themselves, and with Keck KCWI of additional GCs have broadly supported the original interpretation, though the precise dark-matter fraction remains under debate. A reasonable summary as of 2026 is that DF2 and DF4 host extremely little dark matter within their visible bodies, with the precise fraction uncertain at the factor-of-a-few level.
Origin theories — and why they all struggle
No single formation channel explains every UDG, and the heterogeneity of the population is now part of the standard framing. The leading scenarios are sketched here in approximate order of how much of the population each is thought to cover.
| Scenario | Mechanism | Predicts DM | Best evidence |
|---|---|---|---|
| Failed Milky Way | Gas lost early (reionisation, feedback), normal halo remains, very few stars form | Yes — full halo | High GC counts, cluster-rich environments |
| Tidal harassment / stripping | Normal dwarf is puffed up and tidally heated in a cluster; halo is stripped | Some — partial halo | UDG abundance correlates with cluster mass |
| High-spin halos | Unusually high angular momentum spreads the gas to low surface density | Yes — full halo | Predicts large R_eff at fixed M_halo; matches scaling |
| Tidal-dwarf galaxy | Formed from cold debris of a major merger; no halo to inherit | No — no halo | NGC 1052-DF2/DF4; bullet-galaxy debris simulation |
| Feedback-puffed dwarf | Strong supernova feedback expands stellar orbits, lowering surface brightness | Reduced | NIHAO and FIRE simulations reproduce subset |
The high-spin scenario, proposed by Amorisco and Loeb in 2016, has the attractive feature that it predicts a continuum: ultra-diffuse galaxies are the high-spin tail of a normal dwarf population, with no new physics needed. It also predicts a specific scaling between R_eff and the underlying halo spin parameter λ that is, in principle, testable with stacked kinematics.
The tidal-dwarf hypothesis, by contrast, is the only scenario that naturally explains DF2 and DF4. In a major merger, gas in a tidal tail can collapse into a self-gravitating cloud that becomes a real galaxy — but with no inherited dark-matter halo, because the halos of the progenitors are dynamically warm and do not follow the cold gas into the tail. A 2022 paper proposed that DF2 and DF4 were formed in a single "bullet" event in the NGC 1052 group, in which a high-velocity galaxy-galaxy collision separated baryons from their dark-matter halos and left behind a line of dark-matter-free objects. The scenario predicts a chain of similar UDGs along the collision axis, and partial confirmation came when several more candidate dark-matter-deficient UDGs were identified in the predicted positions.
UDGs as a stress test of MOND
Modified Newtonian dynamics (MOND) is a phenomenological alternative to dark matter, originally proposed by Mordehai Milgrom in 1983. It posits that Newton's second law (or, in some formulations, the gravitational law itself) is modified below a fixed acceleration scale a₀ ≈ 1.2 × 10⁻¹⁰ m/s² — above which gravity is normal, below which it is enhanced. MOND has been remarkably successful at reproducing flat rotation curves of disk galaxies without invoking dark matter, and the universality of a₀ across systems has been one of its most striking predictions.
Crucially, MOND is a property of the gravitational law, not of an extra mass component. So in any system where the internal accelerations are below a₀, MOND demands an enhanced kinematic dispersion. NGC 1052-DF2 sits firmly in the MOND regime — its acceleration scale is well below a₀ at its half-light radius. MOND therefore predicts σ_los ≈ 13.4 km/s with no dark matter; the observed σ_los is roughly 8 km/s. That is in tension with MOND at multi-σ.
Proponents have argued that the "external field effect" (EFE) — a feature of MOND in which an external gravitational acceleration suppresses the modification inside a satellite — can be invoked because DF2 is close to its host NGC 1052. With the right EFE strength, the predicted dispersion comes down to consistency with the observed value. Critics counter that DF4, which is comparably distant from NGC 1052 but in a different geometry, would not be similarly suppressed and yet shows the same low dispersion. The MOND debate around UDGs is unresolved and is one of the few empirical tests where dark matter and modified gravity make qualitatively different predictions.
Globular clusters as tracers and as anomalies
The dynamics of NGC 1052-DF2 are inferred from the radial velocities of ten of its globular clusters, not from the stars themselves — the surface brightness is so low that stellar spectroscopy with current 8 m-class telescopes is borderline. GCs are bright, point-like, and individually spectroscopically accessible to the limit of Keck and the VLT, which makes them the only practical kinematic tracer for the faintest UDGs.
This has a downside: GCs are a small-N tracer. Ten clusters give a σ_los estimate with about a 30% statistical uncertainty (Poisson scaling as 1/√N), so the difference between "no dark matter" and "halo a few times more massive than the stars" relies on the tail of the distribution. Different statistical estimators (biweight, MLE, Bayesian) and different anisotropy assumptions give slightly different inferred dispersions, and that range maps to a corresponding range in inferred dark-matter content.
The clusters themselves are anomalous in another respect: NGC 1052-DF2's GC luminosity function peaks roughly 1.5 magnitudes brighter than the universal GCLF turnover seen in every other galaxy. If the GCLF peak is a true standard candle, DF2 is at the distance van Dokkum et al. claim; if the peak shifts in some galaxies, then a closer DF2 — and consequently more dark matter — becomes allowed. The brightness offset is one of the most cited specific anomalies in the literature.
Where UDGs are found
- Coma cluster. The original Dragonfly target, and still the richest known UDG environment, with about a thousand confirmed and candidate UDGs over the ~2-Mpc cluster radius. Coma's UDG distribution shows clear central concentration but extends well into the cluster outskirts.
- Virgo cluster. The closest rich cluster (16 Mpc) and a natural follow-up target; a few hundred candidates known, with the advantage that GC kinematics and even resolved stellar populations are accessible with HST.
- NGC 1052 group. Home of DF2 and DF4 plus several more candidate dark-matter-deficient UDGs; the cleanest known example of a non-cluster UDG environment.
- Local Group. The Antlia 2 dwarf, discovered in 2018 by ESA Gaia, has a UDG-like surface brightness profile and is the closest known example — at ~130 kpc from the Milky Way it is a satellite of our own galaxy. Its proximity makes it the cleanest case for detailed dynamics.
- Field galaxies. Isolated UDGs are known but rare; the discovery rate per unit volume is roughly 1% of the cluster rate, supporting the picture that the dense environments help produce them.
Common pitfalls
- Conflating UDGs with dwarf galaxies. UDGs satisfy a different criterion — size, not luminosity. Many dwarfs are compact at low luminosity; many UDGs are extended at the same luminosity. The class is defined by R_eff and μ₀, not M*.
- Treating "no dark matter" as universal. Only a handful of UDGs (NGC 1052-DF2, DF4, a small number of candidates in NGC 1052 and elsewhere) appear dark-matter-deficient. The Coma UDG population as a whole has standard or even elevated dark-matter content — they were originally inferred to sit in unusually large halos, not unusually small ones.
- Assuming a single origin. The UDG class is operationally defined; it almost certainly contains members produced by several different physical processes. Statements about "the origin of UDGs" should be qualified to a specific subpopulation.
- Confusing surface brightness with total brightness. A UDG has Milky-Way-class total luminosity in the low-mass cases (10⁸ to 10⁹ L☉) — its total brightness is dwarf-like, not Milky-Way-like. What is anomalous is that the light is spread over a Milky-Way-sized area, giving a low surface brightness.
- Reading too much into a single distance. Several specific UDG anomalies — the σ measurement of DF2, the GCLF peak — depend on the assumed distance. A 10–20% revision in distance can substantially change the inferred dark-matter content. Independent distance ladders are critical.
Why they matter
UDGs are interesting in their own right as an unexpected population of large, faint galaxies, and they have substantially extended the size-luminosity plane on which the cosmic galaxy census is laid out. But their largest implication is as an empirical lever on the dark-matter question. In a ΛCDM universe, galaxies and halos co-evolve; the visible matter and the dark matter are dynamically intertwined and almost impossible to separate observationally. NGC 1052-DF2 and DF4, if their interpretation holds, represent the first concrete examples of galaxies that have been separated from their halos — galaxies in which the visible matter and the dark matter are demonstrably distinct, and in which one can be present without the other.
For modified-gravity proponents, the same systems are a problem: a gravitational law cannot be turned off in particular galaxies, so a system that behaves "without dark matter" in the MOND regime is hard to accommodate. Whichever way the dust settles, UDGs are forcing a genuinely new kind of test on the standard cosmological model.
Frequently asked questions
What is an ultra-diffuse galaxy?
An ultra-diffuse galaxy is a galaxy with an effective (half-light) radius greater than about 1.5 kpc — comparable to the Milky Way — but a central surface brightness fainter than about 24 g-band magnitudes per square arcsecond. That combination puts roughly a Milky Way's diameter of stars into the same total brightness as a small dwarf galaxy: about 10⁸ to 10⁹ solar luminosities. The class was formally introduced by van Dokkum and collaborators in 2015 after the Dragonfly Telephoto Array found roughly a thousand examples in the Coma cluster.
How were UDGs discovered if they are so faint?
Standard wide-field surveys are designed for high signal-to-noise on compact sources and tend to flag low-surface-brightness extended structure as flat-field artefacts. Dragonfly is a different instrument: a wide-field array of Canon telephoto camera lenses with nano-fabricated anti-reflection coatings, optimised to image diffuse light without the scattered-light halos that pollute reflecting telescopes. With Dragonfly, van Dokkum et al. (2015) imaged the Coma cluster and identified about a thousand large, faint galaxies that had been overlooked by Sloan and other surveys.
Why is NGC 1052-DF2 so famous?
NGC 1052-DF2 is a UDG in the NGC 1052 group whose globular clusters have radial-velocity dispersions consistent with the galaxy carrying almost no dark matter. Every other galaxy that has had its dynamics measured contains roughly five to ten times as much dark matter as visible matter; DF2 (and its sibling DF4) appear to be the first known exceptions. If the result holds up, it is the first observational evidence that dark matter is a separable component of galaxies — something that can in principle be stripped or removed — rather than an inseparable feature of every system. It also rules out theories in which dark-matter phenomena are really a modification of gravity that acts the same way in every galaxy.
Where do ultra-diffuse galaxies come from?
There is no single accepted origin story. Leading scenarios are: (1) "failed Milky Ways" — galaxies that lost their gas to feedback or reionisation before they could finish forming stars, so they assembled normal halos but ended up with very little luminous mass; (2) tidal stripping in dense environments, where a normal dwarf is puffed up and its halo stripped by interactions inside a cluster; (3) high-spin halos (Amorisco & Loeb 2016), where an unusually high angular momentum spreads the gas out into a low-surface-brightness configuration; (4) tidal-dwarf galaxies formed from debris in a major merger, which can plausibly explain the dark-matter-deficient examples. Most likely the UDG class is heterogeneous and different mechanisms make different members.
Do UDGs have lots of globular clusters?
Many of them, yes. NGC 1052-DF2 hosts roughly 11 luminous globular clusters with brightnesses about a factor of two above the canonical peak of the globular cluster luminosity function — both anomalies. The "too many GCs" observation has been used to argue both for and against a dark-matter origin: in some scenarios GC number correlates with halo mass, in others GCs are produced more efficiently in low-density environments. The Coma UDGs show a wide range in GC number, with the largest hosting more than a hundred.
Is the lack of dark matter consistent with MOND?
No — at least naively. Modified Newtonian dynamics (MOND) predicts an apparent excess of gravity below an acceleration scale a₀ ≈ 1.2 × 10⁻¹⁰ m/s². In a galaxy like NGC 1052-DF2 the internal accelerations are below a₀, so MOND predicts the same large velocity dispersion as a heavy dark-matter halo would. Because DF2's measured dispersion is small, the system is in tension with MOND too. Some MOND advocates have proposed that DF2's external gravity from NGC 1052 itself suppresses the modification (the "external field effect"), and that the result is consistent with MOND given the right environment. The debate is unresolved.
Are UDGs only found in clusters?
They are most abundant in clusters — Coma, Virgo, Fornax — where stripping and harassment are most effective, but isolated UDGs are also known in the field. The cluster preference is one of the strongest hints that environment helps make them, by puffing up otherwise normal dwarfs or by removing their gas, even if not every UDG was made that way.