Simulated cosmic filaments reveal a shared dark-matter pattern

Finding a common scale

Cosmic filaments are elongated concentrations of matter connecting the dense nodes of the cosmic web. They channel matter toward larger structures, but their boundaries and internal structure are harder to define than those of dark-matter halos, the gravitationally bound concentrations associated with galaxies and clusters.

Peng Xu and colleagues investigated whether these strands share a common radial density pattern: how dark-matter density changes with distance from a filament’s central spine. Their study uses the TNG50-1 cosmological simulation, not direct observations. The analysis discussed here is presented in arXiv:2604.05033v2.

The main finding is approximate self-similarity. When radial distances were expressed in units of the connecting node’s virial radius—a characteristic size of its halo—average filament profiles from different epochs became much more alike. This does not establish an exact law for every filament, but it identifies a useful common scale for describing their structure.

First, locate the center correctly

TNG50-1 follows dark matter, gas, stars and black holes in a simulated box 51.7 comoving megaparsecs across. Comoving distances account for cosmic expansion. The team analyzed five snapshots, at redshifts 0, 0.5, 1, 2 and 3, spanning the simulated present back to earlier cosmic epochs.

The researchers reconstructed dark-matter density on a grid with 512 cells along each side, smoothed the logarithmic density field, and used the DisPerSE algorithm to identify filamentary structures. They removed short or unsuitable structures and associated filament maxima with resolved halos. In their catalog, an individual filament runs from a maximum associated with a halo to a saddle point in the density field, rather than necessarily spanning the entire connection between two halos.

A crucial complication emerged: the spines returned directly by DisPerSE did not always coincide with the underlying density ridges. Averaging around an offset axis can make a dense filament center appear artificially diffuse.

The researchers addressed this with a “shrinking-cylinder” procedure. Around each filament segment, they calculated the center of mass inside a cylinder, shifted its axis sideways toward that center, reduced its radius by 10%, and repeated. The iteration stopped when the cylinder contained fewer than 100 sampled particles or the displacement fell below one comoving kiloparsec. Segments were translated without being rotated.

A fixed random sample containing 2% of the simulation’s dark-matter particles was used for both re-centering and profile measurements. The authors report statistically indistinguishable profiles when using a 10% sample instead.

Re-centering increased inferred central densities by as much as 1–1.5 orders of magnitude, roughly a factor of 10–32. This changed the measurement, not the simulated matter distribution. The corrected profiles revealed inner power-law behavior that the original centering had blurred.

Similar profiles, with important exceptions

The team measured density in cylindrical shells, excluded regions within the connecting halos’ virial radii, and combined profiles using length-weighted averages. Densities were expressed relative to the cosmic mean dark-matter density.

Scaling radial distances by the connecting node’s virial radius largely removed differences between epochs, with approximate agreement down to about one-tenth of that radius. Dependence on filament length was negligible in the tested categories, while dependence on node mass was generally weak.

The agreement was not perfect. The all-particle profiles retained some central-density evolution, with lower central densities at higher redshift when node mass was held fixed. Filaments connected to the most massive nodes also showed differences that the authors caution may reflect small samples. No filaments occupied their highest node-mass category at redshift 3.

A generalized triple-power-law formula described the stacked profiles better than the alternative generalized Navarro–Frenk–White model tested. This is an empirical fit, not a physical explanation of the profile. In particular, the authors warn against assigning physical significance to the fitted outer-profile parameters beyond the model’s intended fitting regime.

Clumps account for much of the central peak

To distinguish embedded halos from smoother material, the researchers removed dark-matter particles bound to resolved central halos above 100 million solar masses. They also subtracted the volume those halos occupied, using random sampling to account for overlapping halo regions.

The remaining component developed a flat core within roughly one-tenth of the connecting node’s virial radius. Its central density was up to two orders of magnitude lower than the full profile. This indicates that embedded low-mass halos, rather than a sharply concentrated smooth component alone, produce much of the apparent central cusp.

The smoother component showed the opposite redshift trend from the full profile: it was centrally denser at higher redshift. The authors interpret this as suggesting a shift from smoother filamentary feeding early on toward clumpier feeding later. The density comparison is not itself a direct measurement of accretion rates.

The findings remain limited by one simulation volume, finite resolution and choices in filament identification. The catalog prioritizes clean structures over completeness. Its “smooth” or “unbound” component is defined operationally by removing particles associated with the specified resolved halos; the procedure does not establish the absence of every smaller bound structure.

Applying the result to observations also requires care. Galaxies do not trace dark matter uniformly, and filament profiles reconstructed from galaxies need not match profiles measured directly from simulated dark matter. Larger, equally detailed simulations and reliable reconstruction from galaxy tracers will be needed to test how broadly this common pattern applies.

Source and version details

The original arXiv Atom record and the exact revision’s submission history establish the first submission as April 6, 2026, at 18:00:04 UTC, corresponding to the supplied metadata timestamp of 2026-04-06T18:00:04.000Z. Version 2 was submitted on November 5, 2026, at 22:07:20 UTC. The original-submission timestamp is distinct from the revision date and is not a journal-publication date.

Both arXiv sources retain the comment “12 pages, 8 figures, 2 table”. The supplied complete PDF, however, runs to 13 pages and contains eight figures and two tables, including the appendices. The 13-page count describes the examined document; it does not replace or silently correct the historical arXiv comment. Likewise, the original arXiv abstract is source metadata, separate from this article’s editorial summary, and is not rewritten here.

The arXiv Atom record and landing page also list the journal reference “The Astrophysical Journal Letters, 1009(2), L26 (2026)” and the related journal DOI, 10.3847/2041-8213/aea298. These are authenticated statements in the arXiv record, not details inferred from the PDF’s draft header or independently checked against a publisher record. The PDF header does not negate the journal reference recorded by arXiv.