An Attempt To Compare MOND v. CDM

A preprint finds a draw between MOND and CDM, with each having issues in some domains.

The rotation curves of the 175 disk galaxies in the SPARC database are confronted, one by one and under identical conditions, with the two mass models that have dominated the missing-mass debate for four decades: a cold dark-matter halo of Navarro-Frenk-White (NFW) form, and Milgromian dynamics (MOND) with the acceleration constant held at a0=1.2×10^−10 m s^−2. Stellar mass-to-light ratios, distances and inclinations are treated as nuisance parameters with priors set by the 3.6 μm photometry and by the SPARC error budget, and every galaxy is sampled with a Markov-chain Monte Carlo. 

Three findings emerge. 

(i) Judged by the Bayesian information criterion, the free NFW halo is preferred in 96 galaxies and MOND in 79; once the halo is required to obey the ΛCDM concentration-mass relation the score becomes 86 to 89, i.e. a draw. The two theories fail in different places: MOND loses in massive, high-surface-brightness, bulge-dominated systems, NFW loses in slowly rising dwarf rotation curves. 

(ii) The NFW concentrations demanded by the data sit a median 0.24 dex below the ΛCDM expectation, and 53% of the sample lies more than 2σ below it, independent of halo mass; this is the core-cusp problem expressed as a scaling-relation offset rather than as a handful of anecdotes. 

(iii) The radial acceleration relation built from the full sample has a0= (1.13 ±0 .02) × 10^−10 m s^−2 and an observed scatter of 0.14 dex, yet the value of a0 preferred by individual galaxies spans more than a decade and correlates with surface brightness. The two statements are not in contradiction: the population-level relation is tight because it averages over exactly the degeneracies with mass-to-light ratio and distance that make a0 ill-determined galaxy by galaxy. The full table of fits is provided.

Mikhael Da Silva, "Cusps, cores, and one acceleration: dark-matter haloes versus modified dynamics across the full SPARC sample" arXiv:2610.05871 (October 5, 2026).