The Radial Acceleration Relation (RAR) is a remarkably tight empirical correlation between observed gravitational acceleration (g_obs) and baryonic gravitational acceleration (g_bar) in galaxies, described by a single-parameter function with characteristic scale g† ≈ 1.2 × 10⁻¹⁰ m/s². Established by McGaugh, Lelli & Schombert in 2016 using the SPARC database, the RAR spans galaxies of wildly different masses, sizes, and morphologies with only ~0.13 dex observed scatter — an intrinsic scatter now measured at just 0.034 dex. Whether this extraordinary tightness reflects a fundamental new law of nature (as MOND predicts) or emerges from galaxy formation physics within ΛCDM remains the central unresolved question in extragalactic dynamics. A decade of intense debate has produced rich observational constraints, sophisticated statistical analyses, and discriminating predictions — including "hooks" and "bends" in individual galaxy trajectories — that are beginning to test these frameworks at their limits.
The intellectual roots of the RAR trace to Sanders (1990) and McGaugh (2004, ApJ, 609, 652), who identified the mass discrepancy–acceleration relation (MDAR): the ratio of total-to-baryonic mass in galaxies increases systematically with decreasing acceleration below a critical scale a₀ ≈ 1.2 × 10⁻¹⁰ m/s². The breakthrough came when Lelli, McGaugh & Schombert (2016, AJ, 152, 157) assembled the SPARC database — 175 nearby disk galaxies with homogeneous 3.6 μm Spitzer Space Telescope photometry and high-quality HI/Hα rotation curves. SPARC galaxies span roughly five decades in luminosity, four in surface brightness, and cover morphological types from S0 to irregular dwarfs. The 3.6 μm band was the critical innovation: at this wavelength the stellar mass-to-light ratio (Υ★) is approximately constant at ~0.5 M⊙/L⊙, dramatically reducing the dominant systematic uncertainty that plagued earlier optical studies.
Armed with SPARC, McGaugh, Lelli & Schombert (2016, Physical Review Letters, 117, 201101) plotted g_obs against g_bar for 2,693 resolved data points across 153 galaxies. All points collapsed onto a single relation described by the function g_obs = g_bar / [1 − exp(−√(g_bar/g†))], where the sole free parameter g† = 1.2 × 10⁻¹⁰ m/s². At high accelerations (g_bar ≫ g†), g_obs approaches g_bar — the Newtonian regime. At low accelerations (g_bar ≪ g†), g_obs → √(g_bar · g†) — the deep-MOND regime, predicting flat rotation curves. The observed scatter of 0.13 dex was attributed largely to observational uncertainties in distance and inclination. The authors described this as "tantamount to a natural law for rotating galaxies."
Lelli, McGaugh, Schombert & Pawlowski (2017, ApJ, 836, 152) extended the RAR to 240 galaxies spanning nine decades in stellar mass: 153 late-type spirals and irregulars, 25 early-type galaxies from ATLAS³ᴰ and Chandra, and 62 dwarf spheroidals. All followed the same relation within uncertainties. Residuals showed no correlation with any galaxy property — baryonic mass, gas fraction, effective radius, or morphology — leading the authors to title their paper "One Law to Rule Them All."
The intrinsic scatter of the RAR has become the pivotal battleground between competing theoretical frameworks, because MOND predicts essentially zero intrinsic scatter while ΛCDM simulations typically produce 0.06–0.08 dex.
Li, Lelli, McGaugh & Schombert (2018, A&A, 615, A3) used MCMC to fit the mean RAR to each of the 175 individual SPARC galaxies, marginalizing over Υ★, distance, and inclination with physically motivated Gaussian priors. The residual rms dropped to 0.057 dex. Crucially, a generalized RAR allowing g† to vary galaxy-by-galaxy did not improve fits — no credible evidence for variation in the acceleration scale emerged. The data appeared consistent with a single effective force law. Desmond (2023, MNRAS, 526, 3342) pushed this further using Hamiltonian Monte Carlo to jointly infer RAR parameters and all ~900 galaxy nuisance parameters simultaneously. He measured an intrinsic scatter of just 0.034 ± 0.002 dex, calling the RAR "practically monotonic." Stiskalek & Desmond (2023, MNRAS, 525, 6130) then demonstrated that the RAR is the unique fundamental one-dimensional dynamical correlation for late-type galaxies: no additional galaxy property can tighten it, and all other dynamical scaling laws (Tully-Fisher, Faber-Jackson, baryon-halo conspiracy) derive from it.
Not everyone agrees. Rodrigues, Marra, Del Popolo & Davari (2018, Nature Astronomy, 2, 668) used Bayesian inference with flat priors on a₀ and reported that the acceleration scale varies galaxy-to-galaxy at >10σ, claiming "the RAR is not a fundamental relation." This triggered a vigorous exchange. McGaugh, Li, Lelli & Schombert (2018, Nature Astronomy, 2, 924) argued the flat priors introduced severe parameter degeneracies and that neglecting inclination errors was a critical omission. Kroupa et al. (2018, Nature Astronomy, 2, 925) reinforced these criticisms. Cameron, Angus & Burgess (2020, Nature Astronomy, 4, 132) identified the Rodrigues et al. approach as a fundamentally flawed non-Bayesian test wrapped in Bayesian machinery. Most decisively, Li, Lelli, McGaugh, Schombert & Chae (2021, A&A, 646, L13) performed a reductio ad absurdum: treating Newton's gravitational constant G as a free parameter with flat priors, they "ruled out" G as a universal constant at high confidence — demonstrating the methodology was fatally flawed. Marra, Rodrigues & de Almeida (2020, MNRAS, 494, 2875) updated their analysis with improved priors and quality cuts but maintained a 5.3σ rejection, while the Li et al. critique showed this conclusion remained an artifact of the flat prior on g†.
Stone & Courteau (2019, ApJ, 882, 6) assembled the independent PROBES sample of >2,500 spirals and measured an intrinsic scatter of 0.11 ± 0.02 dex in the stellar RAR. However, their use of heterogeneous optical photometry (rather than 3.6 μm), long-slit Hα spectroscopy, and stellar rather than total baryonic mass introduces additional scatter that likely reflects data quality rather than astrophysical variation. The MIGHTEE-HI study (Vărăşteanu et al., 2025, MNRAS) measured the RAR using 19 HI-selected galaxies out to z ~ 0.08 with resolved stellar masses from multi-band SED fitting, finding intrinsic scatter of 0.045 ± 0.022 dex — consistent with the SPARC-based measurements and providing the first tentative evidence for redshift evolution of the acceleration scale.
Each galaxy traces a trajectory through the g_obs–g_bar plane as one moves from inner to outer radii. Whether these individual tracks faithfully follow the mean RAR, or exhibit systematic deviations that encode dark matter halo structure, has become a frontier area of investigation.
Mercado, Bullock, Moreno et al. (2024, MNRAS, 530, 1349) provided the most comprehensive study, using 20 FIRE-2 zoom-in simulations spanning M★ ≈ 10⁷–10¹¹ M⊙. They identified two key features in individual galaxy tracks. Downward "hooks" appear in low-mass galaxies where stellar feedback has carved cores in dark matter halos: the total acceleration profile becomes non-monotonic, so that moving inward, g_obs can decrease even as g_bar increases, causing the track to curve below the mean RAR. "Bends" appear at very large radii (extremely low accelerations) where g_obs asymptotes to g_bar/f_b (f_b being the cosmic baryon fraction) rather than continuing along the MOND extrapolation. Observed SPARC galaxies DDO 154 and UGC 03546 show clear hook signatures. These features challenge modified-inertia versions of MOND (which predict strictly monotonic tracks) while arising naturally in ΛCDM.
Eriksen, Frandsen & From (2021, A&A, 656, A123) introduced a geometric classification of individual galaxy acceleration curves as "cuspy," "neutral," or "cored" based on trajectory shape. They found that MOND modified-gravity predicts cored geometries for isolated galaxies while MOND modified-inertia predicts neutral geometries — yet observed SPARC galaxies show a mix of all three types, posing a "cusp-core-like challenge for MOND." Li, Lelli, Schombert & McGaugh (2022, ApJ, 927, 198) used semi-empirical models to show that NFW cusps produce upward hooks in low-mass galaxies (where dark matter dominates even at small radii), while adiabatic contraction worsens the discrepancy. Only stellar feedback that flattens cusps can bring models onto the observed RAR.
At the lowest masses, the RAR may break down entirely. Júlio, Read et al. (2025, A&A, 704, A330) studied 12 nearby dwarf galaxies (10⁴ < M_bar/M⊙ < 10⁷·⁵) and found most lie systematically above the extrapolated RAR. Individual dwarfs trace distinct, sometimes multi-valued loci — two different g_obs values for the same g_bar — breaking the assumption of a one-to-one mapping. The pair Draco and Carina, with similar baryonic masses and Milky Way orbits, occupy statistically different positions on the RAR, while Fornax falls below the relation due to its feedback-induced core. EDGE ΛCDM simulations reproduce this behavior through scatter in the M★–M₂₀₀ relation and diverse star formation histories. The connection between rotation curve diversity (Oman et al. 2015, MNRAS, 452, 3650) and RAR trajectories was made explicit by Ghari, Famaey, Laporte & Haghi (2019, A&A, 623, A123), who showed that the diversity of baryon-induced accelerations at 2 kpc drives large rotation curve variation while maintaining a globally tight RAR.
MOND predicted the RAR three decades before its precise observational establishment. Milgrom's original 1983 papers (ApJ, 270, 365/371/384) proposed that dynamics depart from Newtonian behavior below a₀ ≈ 1.2 × 10⁻¹⁰ m/s², immediately predicting the functional form later observed. The RAR equation is essentially the MOND interpolation function, making the relation a natural law within this framework. Famaey & McGaugh (2012, Living Reviews in Relativity, 15, 10) provided the definitive MOND review, showing that a₀ appears consistently across rotation curves, the Tully-Fisher relation, and the central surface density relation. Relativistic extensions include Bekenstein's (2004) TeVeS and the more recent AeST theory by Skordis & Złośnik (2021), which passes cosmological tests including CMB predictions. A key MOND prediction — the external field effect (EFE), violating the strong equivalence principle — was claimed detected at >4σ by Chae, Lelli, Desmond et al. (2020, ApJ, 904, 51; 2021, ApJ, 921, 104) using SPARC galaxies, though Paranjape & Sheth (2022, MNRAS, 517, 130) argued an EFE-like signal can arise in ΛCDM with a different sign. MOND's principal challenges remain galaxy clusters (where it underpredicts mass by a factor of ~2), cosmology, and a tension between RAR-fitting interpolation functions and Solar System constraints from Cassini (Desmond et al. 2024, MNRAS, 530, 1781).
ΛCDM approaches explain the RAR as emergent from galaxy formation physics, with multiple simulation groups reproducing approximate versions. Keller & Wadsley (2017, ApJL, 835, L17) showed 32 MUGS2 galaxies match the SPARC RAR with ~0.06 dex scatter. Navarro et al. (2017, MNRAS, 471, 1841) argued analytically that the RAR arises from the self-similar NFW profile structure plus the stellar-mass–halo-mass relation. Ludlow et al. (2017, PRL, 118, 161103) demonstrated with EAGLE simulations that the relation persists regardless of feedback prescription, though their characteristic acceleration scale (2.6 × 10⁻¹⁰ m/s²) is roughly double the observed value. Paranjape & Sheth (2021, MNRAS, 507, 632) showed quasi-adiabatic relaxation of dark matter reproduces the RAR "without tuning." However, Desmond (2017, MNRAS, 464, 4160) found that fiducial ΛCDM models overpredict the observed scatter by ~3.5σ, even with zero scatter in the stellar-mass–halo-mass relation — suggesting the stochastic hierarchical formation of dark matter halos introduces irreducible scatter that exceeds observations. The Tenneti et al. (2018, MNRAS, 474, 3125) MassiveBlack-II simulation produced a power-law RAR with no clear acceleration scale at all, highlighting that not all ΛCDM implementations reproduce the correct functional form.
Verlinde's emergent gravity (2011, JHEP, 2011, 29; 2017, SciPost Physics, 2, 016) offers a conceptually distinct path, deriving an additional "dark gravitational force" from the volume-law entropy contribution in de Sitter spacetime. For a point mass, emergent gravity recovers MOND phenomenology with a₀ = cH₀/6 ≈ 1.1 × 10⁻¹⁰ m/s² — tantalizing because it connects the acceleration scale to cosmology. Brouwer et al. (2017, MNRAS, 466, 2547) found the framework consistent with galaxy-galaxy weak lensing. However, Lelli, McGaugh & Schombert (2017, MNRAS Letters, 468, L68) showed that for extended mass distributions (real galaxies), emergent gravity departs significantly from both MOND and the observed RAR, predicting radius-dependent residuals that are not seen in the data. The framework remains theoretically promising but observationally challenged at galaxy scales.
Several observational frontiers test the limits of RAR universality. Galaxy clusters show a clear RAR-like correlation but with an acceleration scale ~10× higher than for galaxies (g‡ ≈ 2.0 × 10⁻⁹ m/s²; Tian et al. 2020, ApJ, 896, 70; Chan & Del Popolo 2020, MNRAS, 492, 5865), indicating the galaxy-scale RAR does not extend straightforwardly to cluster scales. Elliptical galaxies mostly follow the same RAR (Chae, Bernardi et al. 2020, ApJL, 903, L31), though slow rotators may show systematic offsets and >5σ deviations have been claimed. Ultra-diffuse galaxies have generated particular excitement: NGC 1052-DF2 and DF4, with very low velocity dispersions suggesting little dark matter, were initially seen as challenging MOND but are now understood to be consistent with MOND's external field effect (Famaey, McGaugh & Milgrom 2018; Haghi et al. 2019, MNRAS, 487, 2441). Freundlich et al. (2022, A&A, 658, A26) found Coma cluster UDGs lie on the RAR and agree with isolated MOND predictions — puzzlingly, even without the expected strong EFE.
The RAR's functional form itself may not be as firmly established as commonly assumed. Desmond, Bartlett & Ferreira (2023, MNRAS, 521, 1817) applied exhaustive symbolic regression to SPARC data and found better-fitting functions than the standard MOND interpolation. The deep-MOND limit (g_obs ∝ √g_bar) is "little evident" in the data, and current dynamical range and uncertainties preclude definitive identification of the asymptotic behavior. The mass-to-light ratio remains a controlled but significant systematic: Schombert, McGaugh & Lelli (2019, MNRAS, 483, 1496; 2022, AJ, 163, 154) demonstrated that Υ★ at 3.6 μm is reliably constrained to ~0.05 dex uncertainty using stellar population models, though the MIGHTEE-HI study found evidence that resolved M/L ratios vary both across and within galaxies. Recent Milky Way studies using Gaia rotation curves (Chan et al. 2024) report a declining rotation curve at R > 20 kpc that may deviate from the universal SPARC RAR, though uncertainties remain large.
Brouwer et al. (2021, A&A, 650, A113) extended the RAR by two decades into the low-acceleration regime using KiDS-1000 weak lensing and found a ≥6σ difference between early- and late-type galaxy RARs at the same stellar mass. This morphology dependence is not predicted by MOND or emergent gravity but can be accommodated by ΛCDM models where the galaxy-halo mass relation depends on formation history — representing one of the few clear observational advantages for the standard cosmological framework.
The RAR stands as one of the tightest and most puzzling empirical correlations in extragalactic astronomy. Its intrinsic scatter of ~0.034 dex makes it more precise than many laboratory measurements and establishes it as the fundamental one-dimensional dynamical correlation for disk galaxies. The relation was predicted by MOND in 1983 and precisely confirmed in 2016 — a remarkable instance of theoretical prediction preceding observational verification by three decades. Yet ΛCDM simulations increasingly reproduce approximate versions of the RAR through the interplay of NFW halo self-similarity, abundance matching, and feedback, though they generically predict more scatter than observed and sometimes the wrong acceleration scale.
The most discriminating recent developments come from studying individual galaxy trajectories. The discovery of "hooks" and "bends" (Mercado et al. 2024) provides concrete, testable predictions that differ between frameworks: feedback-induced dark matter cores produce downward hooks challenging MOND, while the deep low-acceleration regime may reveal bends toward the cosmic baryon fraction rather than the MOND extrapolation. At the lowest masses, the RAR appears to break down entirely (Júlio et al. 2025), with dwarf galaxies showing multi-valued trajectories and significant galaxy-to-galaxy scatter driven by diverse assembly histories. These findings suggest the RAR is not a universal law in the strictest sense but an extraordinarily tight regularity that emerges from — and encodes — the intimate connection between baryonic and dark matter distributions in galaxies. Whether this connection reflects modified dynamics, dark matter physics fine-tuned by feedback, or something else entirely remains one of the most consequential open questions in fundamental physics.