The cosmic web is the interconnected arrangement of matter into filaments, sheets, dense nodes, and underdense voids that characterizes the universe on scales larger than individual galaxies. Its most conspicuous visible tracers are galaxies and galaxy clusters, but most of its gravitating matter is dark matter. Ordinary matter occupies the same broad framework as stars, diffuse gas, and hot plasma, although its distribution is modified by cooling, shocks, and energy released by galaxies.
The web developed as gravity amplified small variations in the density of the early universe. Its geometry records both those initial variations and the subsequent effects of cosmic expansion, matter composition, and dark energy. It is therefore central to cosmology: the same structures that channel gas into galaxies also provide tests of theories about the origin, contents, and evolution of the universe.
The term describes a statistical and dynamical pattern rather than a single bounded object. There is no unique line separating every filament from a sheet, or every void from its surroundings. Different observations and mathematical definitions emphasize different aspects of a continuous, evolving matter distribution.
Geometry, components, and characteristic scales
The cosmic web has four commonly distinguished environments: nodes, filaments, sheets, and voids. Nodes are compact concentrations where several filaments meet, often containing galaxy clusters or massive groups. Filaments are elongated structures connecting dense regions. Sheets, also called walls, extend predominantly in two directions. Voids are relatively underdense regions surrounded by these denser structures. These categories are useful descriptions, but real structures are irregular, branching, and embedded within larger environments.
Their dimensions cover a wide range. Individual filament segments often extend over several to tens of megaparsecs, while connected systems can span hundreds of megaparsecs. Filament widths depend strongly on the tracer and definition, commonly ranging from below a megaparsec to several megaparsecs. Many catalogued voids have effective radii of roughly 10–50 megaparsecs, although smaller and larger examples occur. A megaparsec is approximately 3.26 million light-years. Cosmologists commonly quote comoving distances, which factor out the average expansion of the universe, rather than distances measured with a physical ruler at the epoch being studied.
Most cosmic volume belongs to low-density environments, whereas a disproportionate share of matter lies in the comparatively small volume occupied by filaments and nodes. Exact volume and mass fractions are not universal constants. They change with the resolution of a map, the scale over which density is smoothed, and the classification method. A tenuous bridge between clusters may qualify as a filament under one definition but merge into a broader sheet under another. Likewise, a large void may contain smaller voids, low-mass galaxies, and faint internal filaments.
The cosmic web does not imply that the universe remains strongly structured on every possible scale. In the standard cosmological description, sufficiently large volumes approach statistical homogeneity and isotropy: averages become approximately independent of location and direction. This does not mean that every large region looks identical, or that exceptionally extended structures cannot exist. It means that statistical fluctuations become smaller as the averaging scale increases. Establishing this transition requires accounting for survey boundaries, selection effects, and the correlations expected in a finite sample.
Nor is every connected overdensity gravitationally bound. A supercluster may consist of clusters, groups, and intervening filaments that share a large-scale density enhancement without being destined to collapse into one object. Some dense cores can be bound while their outer regions continue to separate with cosmic expansion. The web is therefore not a static lattice of solid connections; it is an evolving distribution of matter whose different parts can have very different dynamical futures.
Discovery through galaxy mapping
Evidence for large-scale organization accumulated gradually because galaxies were first observed mainly as positions on the sky. Angular maps revealed concentrations and apparent gaps, but projection mixed objects at different distances. Converting these maps into a three-dimensional picture required measurements of redshift. For sufficiently distant nearby galaxies, the average relationship between redshift and distance reflects cosmic expansion, although local gravitational motions introduce important distortions. The development of efficient spectroscopy made extensive three-dimensional mapping possible.
In 1934, Edwin Hubble examined the distribution of galaxies using photographic counts, while later work by Gérard de Vaucouleurs emphasized the flattened arrangement of nearby galaxies associated with the Local Supercluster. George Abell’s 1958 catalogue of rich galaxy clusters provided another route to tracing structure across large distances. These studies established that galaxies and clusters were not distributed independently, but early catalogues could not yet resolve the web with the completeness of modern surveys.
A decisive change came during the 1970s and 1980s. Redshift studies of the region around the Coma cluster revealed substantial underdense volumes between galaxy concentrations. Work by Stephen Gregory and Laird Thompson, and by Jaan Einasto and collaborators, helped establish the importance of cellular and filamentary organization. The approximately 60-megaparsec-wide Boötes void, reported by Robert Kirshner and colleagues in 1981, provided a striking example of a large region containing far fewer bright galaxies than its surroundings.
The Center for Astrophysics redshift surveys made the geometry especially vivid. A thin survey slice published in 1986 by Valérie de Lapparent, Margaret Geller, and John Huchra displayed galaxies arranged along apparent surfaces surrounding relatively empty regions. In 1989, Geller and Huchra reported the structure now called the CfA Great Wall, extending roughly 150 megaparsecs in the mapped region. Such discoveries challenged simple pictures in which galaxy clusters were isolated islands within an otherwise nearly featureless distribution. They did not, by themselves, establish a violation of large-scale statistical homogeneity.
Subsequent surveys greatly increased both volume and sampling density. The Sloan Digital Sky Survey and the Two-degree Field Galaxy Redshift Survey mapped hundreds of thousands and eventually millions of objects, allowing filament networks, void populations, and clustering statistics to be measured systematically. The expression “cosmic web” became firmly associated with the gravitationally generated network, notably through work by J. Richard Bond, Lev Kofman, and Dmitry Pogosyan in 1996. The modern picture arose from the convergence of these observations with numerical simulations, rather than from a single discovery or image.
Growth from primordial density fluctuations
The standard account begins with small departures from uniform density in the early universe. Their statistical properties are measured particularly well through the cosmic microwave background, whose temperature anisotropies are typically of order one part in one hundred thousand. In models involving cosmic inflation, primordial fluctuations originate from quantum fluctuations stretched to astronomical scales. Inflation supplies a framework for their origin, while the precise physical mechanism remains a subject of research.
A convenient measure of structure is the density contrast,
where is the local matter density and its cosmic mean. In an overdense region, the gravitational attraction exceeds that associated with the average background. Matter develops motions toward the overdensity, increasing its contrast. Underdense regions lose matter relative to their surroundings and become more rarefied. Both processes occur within an expanding universe, so an overdensity can initially grow in contrast even while its physical size is still increasing.
For pressureless matter on appropriate subhorizon scales, linear perturbation theory gives the approximate growth equation
Here is the expansion rate, the gravitational constant, and the mean matter density. The term involving slows growth, while self-gravity drives it. In a matter-dominated universe with negligible spatial curvature, the growing solution is proportional to the scale factor. During radiation domination, and later when dark energy becomes dynamically important, the growth history differs. Small-scale growth also depends on the ability of different matter species to cluster.
Collapse is generally directional rather than spherical. In the approximation introduced by Yakov Zel’dovich in 1970, matter trajectories respond to an initial displacement field whose deformation differs along three principal axes. A region can contract first along one axis, producing a sheetlike configuration; contraction along a second creates a filament; contraction along all three produces a compact concentration. The surrounding gravitational tidal field helps determine these preferred directions. This sequence explains why elongated and flattened structures arise naturally from gravity without requiring a pre-existing network.
The sequence is not a rigid timetable followed by every region. Small objects can form within larger filaments while those filaments are still developing, and mergers continually rearrange nodes and bridges. Once trajectories cross, several streams of collisionless matter can occupy the same position with different velocities, beyond the simplest approximation. Numerical calculations are then needed to follow the resulting nonlinear evolution. The observed web combines hierarchical assembly on small scales with coherent, anisotropic motions across much larger regions.
Dark matter, gas, and the thermal structure of filaments
In the Lambda–cold dark matter model, cold dark matter supplies most of the matter responsible for the web’s gravitational structure. Cosmological measurements indicate that ordinary baryonic matter accounts for approximately 5 percent of the present cosmic energy budget, compared with about 26 percent in dark matter; the remainder is predominantly dark energy. These percentages refer to the total energy budget, not just matter. Because dark matter is effectively collisionless on the scales normally modeled, its streams can interpenetrate rather than forming ordinary fluid shocks.
Gas behaves differently. It responds to pressure, can radiate energy, and can be heated when converging flows create shocks. Falling toward sheets, filaments, and clusters converts gravitationally acquired motion into thermal energy. Much of the low-redshift intergalactic gas is predicted to occupy the warm–hot intergalactic medium, conventionally associated with temperatures around – kelvin. Cooler photoionized gas occupies less strongly heated regions, while the interiors of massive clusters can reach temperatures above kelvin.
This thermal diversity underlies the “missing baryons” problem. The amount of ordinary matter inferred from early-universe measurements exceeds the amount easily counted in nearby stars, cold gas, and bright hot atmospheres. The discrepancy does not mean that baryons disappeared. A substantial fraction is expected in diffuse ionized material whose low density makes it difficult to detect. Absorption lines, X-ray emission, microwave scattering measurements, and radio-burst dispersion provide complementary evidence for this extended reservoir, but assigning a precise fraction specifically to filaments depends on definitions and observational modeling.
Gas and dark matter therefore trace broadly similar structures without sharing identical profiles. Gas pressure smooths small-scale density variations. Radiative cooling can concentrate gas into galaxies, while stellar explosions and energy released around black holes can expel it into surrounding space. These processes redistribute baryons over distances extending beyond galaxy boundaries. In some filaments, the gas temperature, density, and chemical composition are more informative about accretion and past energy injection than the galaxy count alone.
The web also contains heavy elements and potentially dynamically relevant magnetic fields. Oxygen and other elements detected outside galaxies indicate that enriched material has been transported from sites of stellar nucleosynthesis. Proposed magnetic-field origins include weak primordial seeds and later injection or amplification associated with galaxies and plasma motions. Field strengths and coherence lengths in tenuous intergalactic filaments remain difficult to establish. Radio observations constrain combinations of magnetic fields, relativistic particles, and thermal electrons, so interpreting a signal requires separating these contributions rather than treating filament magnetism as directly mapped everywhere.
How the web is observed
The most extensive three-dimensional maps use galaxies as tracers. Spectroscopic surveys measure angular positions and redshifts, then convert them into positions using an assumed cosmological model. The observed radial coordinate is not a pure distance indicator: a galaxy’s motion relative to the average expansion contributes to its redshift. Random motions within clusters create elongated “fingers of God,” while coherent infall produces larger-scale anisotropies. These effects complicate the appearance of filaments but also contain information about gravitational growth.
Galaxies are also biased tracers: different galaxy populations do not follow the matter distribution in the same way. Massive, luminous, or quiescent galaxies tend to be more strongly clustered than many lower-mass, star-forming systems. A bright-galaxy catalogue can therefore make a region look emptier than a deeper sample would. Survey completeness, obscuration, unsuccessful redshift measurements, and the survey footprint all require explicit treatment. A convincing web map is not simply a plot of detected objects; it includes an account of which objects could have been detected.
Gravitational lensing provides a complementary probe because it responds to gravitating matter rather than emitted light. Weak lensing measures small, statistically coherent distortions in background galaxy images. Individual filaments usually produce weak signals, so studies often combine many similarly selected systems. Lensing reconstructions of the cosmic microwave background extend the method to integrated matter distributions across a broad range of distances. Both approaches involve projection, and converting an apparent bridge into a physical filament requires careful control of unrelated foreground and background matter.
Diffuse gas can be examined through absorption against bright background sources. The Lyman-alpha forest consists of numerous absorption features produced by intervening neutral hydrogen in quasar spectra. At redshifts around two to three, much of this absorption traces mildly overdense intergalactic gas rather than isolated clouds with sharp boundaries. Closely spaced sightlines can support three-dimensional tomographic reconstructions. Absorption by more highly ionized elements, including oxygen, probes hotter material, although weak signals and uncertain ionization conditions complicate estimates of total gas mass.
Other methods emphasize different physical quantities. The Sunyaev–Zel’dovich effect measures changes to the microwave background caused by scattering from free electrons; its thermal component depends on electron pressure integrated along the line of sight. X-ray emission is especially sensitive to denser hot gas because its emissivity generally depends strongly on density. Fast radio bursts provide electron-column measurements through their frequency-dependent propagation delays. Fluorescent Lyman-alpha emission can reveal gas near powerful illuminating sources, but such images need not represent typical filaments. Combining these methods helps distinguish matter density, temperature, ionization state, and illumination.
Mathematical identification and statistical description
There is no single universally accepted algorithm for extracting the cosmic web from a set of galaxies or a simulated density field. Some methods begin with local density, identifying maxima as nodes and tracing ridges between them. Others use velocities, gravitational deformation, or topological relations among critical points. Each method answers a somewhat different question: where matter is concentrated, where matter is flowing, or how structures are connected. Differences between catalogues are therefore not necessarily evidence that one has failed.
A widely used dynamical approach examines the eigenvalues of a tensor derived from the gravitational potential. With a suitable sign convention and threshold, three collapsing directions identify a node, two a filament, one a sheet, and none a voidlike environment. Closely related methods use the velocity-shear tensor. Their classifications depend on smoothing scale and threshold: a galaxy can occupy a small filament that lies within a much larger sheet. Consequently, a web label is meaningful only together with the scale and operational definition used to assign it.
Topological methods identify maxima, minima, and saddle points of a reconstructed field, then connect selected critical points into a skeleton. The DisPerSE algorithm, introduced by Thierry Sousbie in 2011, is a prominent example. It uses ideas from discrete Morse theory and topological persistence to separate more significant features from fluctuations associated with sampling noise. Other methods use marked point processes, minimum spanning trees, or multiscale filters. A mathematically thin skeleton should not be mistaken for a direct measurement of a filament’s physical boundary.
Conventional clustering statistics complement these geometric descriptions. The two-point correlation function measures the excess probability of finding pairs at a specified separation, while the matter power spectrum expresses fluctuation strength as a function of spatial frequency. These quantities efficiently summarize scale-dependent clustering but do not uniquely specify the web’s appearance: fields with different phase arrangements can share the same power spectrum. Higher-order correlations, counts in cells, void statistics, and topological measures retain additional information about nonlinear structure and connectivity.
Connectivity itself can be quantified by counting the filaments attached to a node, measuring their lengths and curvature, or recording density profiles around their axes. Void catalogues similarly distinguish an effective radius, a central density, a surrounding density profile, and whether neighboring depressions are merged. Statistical comparisons require matching these definitions between observations and simulations. A reported change in filament abundance can otherwise arise from altered galaxy sampling or a different smoothing scale rather than genuine physical evolution. Uncertainty estimates must also include correlated fluctuations across the finite volume surveyed.
Numerical simulations and reconstruction
Computer simulations connect the initial density field to the mature cosmic web. In a cosmological N-body simulation, numerical particles represent samples of the matter distribution rather than individual physical dark matter particles. Their gravitational interactions are evolved within an expanding background. Periodic boundary conditions commonly approximate a representative volume without artificial edges, while Fourier-based mesh methods and hierarchical tree methods make the force calculation feasible for very large particle numbers.
Early calculations already showed how gravitational instability could generate sheets, filaments, and voids. As computational power increased, simulations became large enough to compare detailed clustering patterns with redshift surveys. The Millennium Simulation, published in 2005, followed more than ten billion dark matter particles in a cube with a side length of megaparsecs, where expresses the Hubble constant in units of kilometres per second per megaparsec. Such calculations demonstrated how a web of dark matter could host hierarchically assembled populations of galaxies and clusters.
Hydrodynamic simulations add gas dynamics, radiative cooling, chemical enrichment, and approximate treatments of star formation and black-hole activity. Projects such as Illustris, EAGLE, and IllustrisTNG use these ingredients to investigate how galaxies and diffuse gas inhabit the underlying network. Processes below the numerical resolution must be represented through subgrid prescriptions. These are physically motivated models rather than direct calculations of every star-forming cloud or black-hole accretion flow, and different prescriptions can produce different gas distributions even when the broad dark matter web is similar.
Every simulation balances volume against resolution. A very large box samples rare massive structures and long-wavelength fluctuations, but resolving small galaxies throughout that volume is expensive. A small, high-resolution calculation follows local physics in greater detail but may miss influential large-scale environments. Zoom simulations combine a highly resolved target region with a coarser surrounding volume. Numerical convergence tests, comparisons among codes, and controlled changes of physical assumptions are necessary to distinguish robust web properties from computational artifacts.
Simulations also enable reconstruction rather than prediction alone. In constrained realizations, observed galaxy distributions or peculiar velocities help infer plausible initial conditions for the nearby universe. Forward-modeling methods evolve candidate initial fields and compare their predicted tracers with survey data, producing probabilistic reconstructions of density and flow. Mock catalogues reproduce survey geometry, selection, redshift distortions, and measurement errors. These are essential for testing whether an apparent structure would be recovered reliably and for avoiding comparisons between an idealized complete simulation and an incomplete observational sample.
Effects on galaxies and their evolution
The cosmic web supplies the environment in which galaxies acquire matter and move. Gas can flow from low-density regions into sheets and filaments, then toward denser nodes and the dark matter halos surrounding galaxies. This does not imply that all inflow follows a single uninterrupted stream. Gas may be shocked, mixed, recycled after ejection, or temporarily stored in the extended atmosphere surrounding a galaxy. The route depends on halo mass, redshift, gas temperature, and the history of energy injection.
Simulations distinguish relatively cool accretion from gas that is strongly heated near a halo’s characteristic gravitational temperature. At high redshift, dense cold streams can penetrate hot surroundings in some massive halos. At later times and in other mass ranges, stable shocks and hot atmospheres can alter the supply of gas available for star formation. Filaments therefore act both as supply routes and as environments where gas is processed before reaching a galaxy. A filament is not necessarily cold throughout: its inner regions can contain shock-heated material alongside cooler components.
Galaxy properties vary with environment. Dense regions contain larger proportions of galaxies with little ongoing star formation, whereas many lower-density regions contain more actively star-forming systems. The classic morphology–density relation, quantified by Alan Dressler in 1980 for cluster environments, concerns the changing proportions of galaxy types with local density. Extending such results to the cosmic web requires separating several correlated influences. Galaxies near filament intersections tend to inhabit different halo populations from galaxies far from nodes, so a filament association alone does not establish a direct filament-specific mechanism.
Processes that change galaxies include mergers, gravitational encounters, loss of gas supply, and removal of gas by motion through surrounding plasma. Their importance differs between clusters, groups, and more tenuous filaments. Some galaxies are “preprocessed” within groups before entering clusters, meaning that part of their evolution occurs before arrival in the final dense environment. Gas surrounding filaments may also affect low-mass halos, but quantifying this influence requires controlling for halo mass and formation history rather than attributing every environmental trend to the same process.
The web also supplies preferred directions. Galaxy shapes, halo orientations, and angular momenta can correlate statistically with nearby filaments or sheets. Simulations often predict changes in alignment with halo mass, merger history, and epoch; observational measurements depend on how galaxy spin and filament direction are estimated. These effects are important beyond galaxy formation because coherent intrinsic alignments can contaminate weak-lensing measurements. A complete interpretation must distinguish alignments generated by gravitational history from distortions caused by light propagation through foreground matter.
Cosmological tests and unresolved questions
The cosmic web provides many tests of the Big Bang framework and the physical ingredients of cosmic evolution. Its statistics depend on the primordial fluctuation spectrum, the amount and properties of matter, and the expansion history. Baryon acoustic oscillations leave a preferred separation in galaxy clustering, inherited from sound waves in the early photon–baryon plasma. The associated comoving scale is about 147 megaparsecs in standard cosmology. Measuring its apparent size across redshift provides a geometric probe distinct from identifying individual walls or filaments.
The rate at which structure grows tests gravity on large scales. Redshift-space distortions constrain combinations involving the growth rate and the amplitude of matter fluctuations. Gravitational lensing measures the integrated influence of matter on light, while galaxy motions respond to gravitational potentials through their dynamics. Comparing these probes can test general relativity and alternatives that modify the relation between matter, motion, and light deflection. Such tests require accurate models of tracer bias, nonlinear evolution, and observational systematics; an imperfect fit does not automatically identify a failure of gravity.
Matter properties also affect the web. Massive neutrinos move rapidly enough to suppress clustering below characteristic scales. Dark matter with a significant primordial velocity dispersion would reduce small-scale structure relative to cold dark matter, while self-interactions could change halo interiors and, in some models, other aspects of structure formation. Filaments, voids, and halo populations can supply complementary constraints, but many proposed signals are partly degenerate with ordinary gas physics, galaxy selection, or changes in the initial fluctuation spectrum.
Voids are useful because their low densities emphasize environments unlike the interiors of collapsed halos. Their abundance, shapes, expansion, and surrounding velocity fields depend on cosmological assumptions. Some modified-gravity theories predict relatively strong departures from general relativity in low-density regions where screening mechanisms are less effective. Translating this sensitivity into reliable constraints is difficult: void boundaries depend on the tracer sample, and sparse sampling can merge distinct underdensities or create apparently empty regions.
Several questions concern the web’s ordinary matter more directly: the detailed partition of baryons among thermal phases, the strength and origin of intergalactic magnetic fields, and the extent to which galaxy activity changes filament gas. Larger redshift surveys, deeper lensing maps, improved X-ray spectroscopy, and denser grids of background absorption sightlines address different parts of this problem. Their interpretation increasingly relies on joint analyses that preserve the distinct sensitivity of each measurement. A gas-pressure map, a galaxy skeleton, and a lensing reconstruction can describe the same region without tracing precisely the same material or boundaries.