Abstract
Cosmic memory cannot be reduced to the general fact that the past contributes to producing the present. This article distinguishes three regimes. The first corresponds to trajectory dependence. The
second appears when present structures preserve reconstructible historical information. It comprises a retrodictive dimension, illustrated by chemical memory and galactic archaeology, in which certain past events can be inferred from present observables, and a more demanding predictive dimension, in which history still supplies information beyond the variables adopted to describe the present state. Galactic assembly bias puts this second dimension directly to the test. The third regime appears when inherited traces are remobilised within new functional cycles, as in the living. The boundary between historical dependence and reconstructibility depends in part on descriptive resolution, while the passage to functional reuse requires an additional organisational architecture. The continuity between cosmic and biological evolution is operative without implying a structural identity.
From accretion memory to a cosmic question
Accretion memory is probably the most immediately visible form of material memory. Growth proceeds mainly by the addition of matter. Earlier stages are not entirely replaced and part of the chronology remains spatially ordered within the present structure. New layers are not deposited in a neutral space. They extend an already constituted form, inherit its geometry and contribute to modifying it. A shell, a coral, a stromatolite, a growth ring or an otolith thus render part of the past directly visible in their present organisation. Older structures become the material support of the new ones. Their persistence does not mean that everything is preserved. Some information is compressed, transformed or lost, but part of the history remains integrated into the form produced.
The Universe does not grow in this way on the global scale. Its expansion does not correspond to the addition of layers around a centre, and its evolution leaves behind no universal succession of intact strata. Yet a more general property brings cosmic evolution close to this logic. New structures appear within an already transformed real. A galaxy does not form from matter without a history. It emerges within an already structured distribution of matter, receives gas flows oriented by the cosmic web, crosses particular gravitational environments and is composed of matter whose chemistry has been modified by earlier generations of stars. New structures inherit conditions that are not produced solely by the local laws acting at the moment of their appearance. They also depend on the transformations that shaped the medium in which those laws continue to operate.
This intuition must not, however, lead us to call memory every consequence of the past. An earlier event almost always modifies the later state of a dynamic system. Formulated without an additional criterion, the principle that the past transforms future possibilities becomes too general to distinguish a memory from mere causal continuity. Scientific content begins when one seeks to determine what persists, in what form, with what legibility and by what criteria that persistence could be falsified.
A Universe that structures itself within an already transformed real. This article mainly uses the standard ΛCDM cosmological model as a descriptive framework for structure formation. This choice reflects the fact that a large part of cosmological simulations, halo models and contemporary analyses of the relation between galaxies and dark matter are formulated within that framework. The notion of cosmic memory developed here does not,
however, suppose that ΛCDM necessarily constitutes the definitive description of cosmological evolution.
In the standard account of structure formation, the small density fluctuations observable in the cosmic microwave background constitute the seeds from which gravitation progressively amplifies matter contrasts. This evolution produces a large-scale organisation composed of voids, sheets, filaments and nodes. The cosmic web is not merely the result of an earlier history. Once constituted, it orients the movements and transformations that follow. Matter joining a halo does not fall into a neutral space. It follows trajectories influenced by already formed matter distributions and gravitational potentials. Flows converge preferentially toward filaments and then toward the densest regions. Halos thus develop within environments whose geometry and dynamics have already been modified by the evolution of neighbouring structures.
Hydrodynamic simulations such as IllustrisTNG show that the relation between galaxies and halos does not necessarily reduce to the halo's present mass. Secondary properties such as concentration, environment or certain assembly-related characteristics can be associated with variations in galactic occupation. Central galaxies and satellites do not always display the same dependences, which reflects the diversity of the trajectories by which they were constituted. Two halos of comparable mass may have gone through different histories of accretion, merger, star formation and feedback. These differences can leave durable effects in the morphology of the galaxies, their gas content, their stellar populations, their star-formation activity or their chemical composition. The state observed today is not necessarily summarised by a single global variable.
This dependence does not mean that every event remains identifiable. A merger can deform or destroy a disc, redistribute stars, heat gases and mix previously distinct populations. Stellar explosions disperse matter. Successive gas inflows can dilute certain chemical signatures. Gravitational interactions can erase coherent structures or transform them into diffuse distributions. The past is neither preserved in full nor ordered like an intact archive. It may be incorporated, mixed, dispersed or rendered hard to access. A trace does not, however, need to preserve the initial form of an event in order to prolong its effects. It can persist as a composition, a statistical distribution, an anisotropy, a stellar population, a gradient or a new constraint. Part of the history then becomes inseparable from the present structure.
Why the influence of the past does not suffice to define a memory. Every non-trivial dynamic system has a history. A star exists because a gas cloud collapsed. A galaxy has its present mass because it accumulated matter. A halo carries the consequences of the interactions that modified its structure. The past necessarily takes part in producing the present. That property nevertheless remains insufficient to define a memory in the strong sense.
In a Markovian description, the present state contains all the information needed for the probabilistic determination of the next state. The past contributed to producing that state, but it is no longer necessary to know the earlier states explicitly in order to describe future evolution. The present state screens them off. A system can be historically produced without displaying an additional dependence on its history once its complete state is known. Conversely, a reduced
macroscopic description may omit variables in which part of that history remains inscribed. What appears as a memory then depends on the level at which the system is observed and on the variables used to represent it.
Several regimes must be distinguished. The first corresponds to trajectory dependence, in which the past contributed to producing the present configuration. The second corresponds to reconstructible memory, in which certain present properties preserve information allowing earlier events or trajectories to be inferred. The third corresponds to functional reuse, in which certain inherited traces are remobilised by the organisation within new cycles of maintenance, development, adaptation or reproduction. These regimes share an operative continuity, since in each of them a past transformation modifies the conditions of subsequent transformations. They nevertheless have neither the same empirical content nor the same organisational architecture.
First regime: trajectory dependence
In the first regime, the present state results from a particular succession of transformations. A merger modifies the distribution of stars. A feedback episode transforms the gas content of a
galaxy. A phase of accretion modifies the mass and dynamics of a halo. A generation of stars transforms the composition of the medium from which others will be able to form. The past takes part in producing the present, but this property remains very general.
Similar macroscopic states may have been reached through different histories. Some of these differences continue to produce effects, while others become practically indiscernible. Trajectory dependence thus constitutes the minimal level of historical inheritance. It indicates that the system cannot be understood as if it appeared at each instant on a blank slate, but it does not demonstrate that a singular event can be reconstructed from the present. When all the relevant variables are included, history may no longer supply additional information. The system remains historically constituted, but its present state suffices to describe its future evolution.
Speaking of memory in this first regime remains possible in a weak sense. The cosmic web, matter distributions, chemical compositions and gravitational structures constitute inherited conditions that orient later transformations, but the mere existence of these constraints does not yet allow a legible memory to be distinguished from ordinary causal continuity. The passage to a more demanding regime depends on the persistence of identifiable historical information.
Second regime: legible or reconstructible memory
The second regime appears when part of the history remains inscribed in present properties in a sufficiently structured way to allow the inference of earlier events, sequences or conditions. It is no longer enough to assert that the past produced the present. One must show that present observables make it possible to distinguish several possible histories or to reduce the set of trajectories compatible with the observed state.
A stellar population may preserve relations between chemical abundances, ages and motions that reveal a common origin. A stellar stream may allow the disruption of an earlier structure to be reconstructed. An orbital distribution may carry the signature of an old accretion event. A chemical composition may indicate the relative rhythms at which different generations of stars enriched their environment. The trace then becomes historically
informative.
The second regime does not, however, correspond to a single empirical property. Two questions must be separated. The first is retrodictive and asks whether present properties allow part of the past to be reconstructed. The second is predictive and asks whether knowledge of that history still supplies information beyond the present variables used to describe the system. These questions are linked, but they are not equivalent. Chemical memory and galactic archaeology bear mainly on the reconstructibility of the past from the present. Research on assembly bias examines a more demanding claim, that of a residual explanatory or predictive value of history.
Chemical memory: a history incorporated into matter. Chemical composition provides one of the most accessible examples of cosmic historical memory. The first generations of stars formed in a medium composed mainly of hydrogen and helium. Their evolution progressively produced and dispersed heavier elements. The following generations were constituted within matter already transformed by the activity of earlier generations.
This enrichment is neither uniform nor linear. It depends on local rhythms of star formation, on stellar masses, on supernova explosions, on stellar winds, on inflows of less enriched gas, on ejections out of galaxies and on merger episodes. Two regions observed at the same epoch may display different compositions because they followed different trajectories. Metallicity does not function as a simple clock. It constitutes a historical property whose interpretation requires reconstructing the processes that contributed to producing it.
The different sources of enrichment do not release the same elements on the same timescales. Massive stars rapidly transform their environment, while other sources contribute with longer delays. The relative abundances observed in stars can thus preserve information about ancient rhythms of star formation and about the populations that preceded their birth. A star never contains a complete copy of the history of its environment, but its composition can preserve certain differences produced by that history.
Observations of JADES-GS-z14-0 show that this enrichment began very early. The detection of the [O III] line at a redshift of 14.1793 indicates that a substantial production of heavy elements had already taken place about three hundred million years after the Big Bang. Metallicity estimates depend on the models employed, but they show that a rapid chemical evolution was already under way in this very ancient galaxy. This observation does not mean that the Universe follows a necessary progression toward complexity. It shows that one generation of structures can transform the matter available to the following generations and that this transformation can remain observable in their composition. Chemical memory is incorporated into the matter of which new structures are made.
Retrodictive reconstructibility: galactic archaeology. Galactic archaeology exploits the possibility of reading some of these traces. It combines the motions, orbits, ages and chemical compositions of stars in order to reconstruct episodes of formation and assembly that are no longer directly observable. Stars
then become material witnesses, not because they would preserve a complete representation of the past, but because their present properties remain bound up with the environments and events that contributed to their formation.
Gaia-Sausage-Enceladus is one of the most emblematic examples. The progenitor galaxy no longer subsists as an independent structure. Its stars were dispersed and incorporated into the halo of the Milky Way. Yet some populations display orbital, kinematic and chemical properties distinctive enough to be linked to an ancient major accretion event. Work devoted to this structure has associated a significant part of the inner stellar halo with that ancient merger and has shown that its consequences can still be studied in the present distribution of stars.
Memory does not lie in the conservation of the vanished galaxy. It lies in the persistence of relations between distributed traces. Present-day stars do not reproduce the event, but they reduce the set of histories compatible with the observations. Models then make it possible to estimate certain properties of the absorbed structure, of its trajectory and of its influence on galactic evolution. The details remain revisable. The exact limits of the populations associated with Gaia-Sausage-Enceladus, the chronology of the merger, its mass and its effects on the various components of the Milky Way continue to be studied. This revision does not call the principle of reconstructibility into question. It shows its real conditions. A material memory is not necessarily a transparent archive. It must be interpreted from incomplete observables, models and comparisons between several possible histories.
Gaia-Sausage-Enceladus thus represents a retrodictive memory. An ancient transformation remains partly accessible through the organisation of the traces it left. This property is distinct from mere trajectory dependence, since it is no longer only a matter of noting that the merger produced consequences. It becomes possible to infer the existence of the event from relations still present.
Residual historical information and predictive sufficiency. A further requirement appears when one asks whether knowledge of the history improves description or prediction beyond a defined set of present variables. Suppose a halo is described solely by its present mass. If that variable summarised all the information relevant to the galactic properties studied, halos of the same mass should display comparable occupations and statistical behaviours, independently of their formation trajectory. Knowing their assembly history would then add nothing.
This question can be formulated in terms of predictive sufficiency. Let G denote the galactic properties studied, M the present mass of the halo and H its assembly history. To say that mass suffices to describe the properties considered amounts to supposing that they are conditionally independent of history once mass is known:
P(G | M, H) = P(G | M)
Once the halo's mass is taken into account, knowing its assembly history should no longer modify the predictions. Within this framework, mass constitutes a predictively sufficient description for the observables studied. If adding the history, or a variable that preserves a reliable signature of it, still modifies the predicted distribution of galactic properties, conditional independence fails and the macroscopic state adopted is not sufficient.
Relative predictive memory can thus be defined as the persistence of historical information not screened off by the present variables used in the model. This formulation does not suppose that the fundamental dynamics of the Universe is non-Markovian. It indicates that the present representation chosen does not summarise all the information relevant to the observables or dynamics considered. A more complete description could conceivably incorporate that historical information into an enriched present state. Predictive memory remains relative to the choice of variables and to the level of description.
This formalisation also makes it possible to distinguish the two dimensions of the second regime more sharply. Retrodictive reconstructibility concerns the information the present preserves about the past. It does not suppose that a present state determines a unique history, nor that an exact inversion of the dynamics is possible. Several trajectories may produce similar states and the reconstruction often remains probabilistic. Predictive sufficiency poses another question. It asks whether the present state adopted screens off history for the prediction of the properties considered. A state may be rich in reconstructible traces while being predictively sufficient. Conversely, a historical variable may improve a prediction without allowing a singular event to be reconstructed precisely.
Assembly bias as a test of relative predictive memory. Galactic assembly bias constitutes a privileged terrain for examining this second dimension. In the simplest models of the relation between galaxies and halos, halo mass constitutes the principal variable for estimating the number and certain properties of the galaxies it contains. If that description were entirely sufficient, halos of the same mass should produce comparable occupation and clustering statistics.
Simulations show, however, that secondary properties can modify this relation. Concentration, formation time, environment and other characteristics may be associated with variations of clustering or galactic occupation at comparable mass. The term assembly bias nevertheless covers several phenomena that must not be conflated. A dependence on present concentration is not necessarily a direct measure of history. A dependence on environment may reflect several correlated mechanisms. Some secondary properties are only imperfect indicators of the assembly trajectory. Interpretation in terms of memory requires specifying what the variable represents and what it adds beyond the variables already included in the model.
Assembly bias becomes a test of relative predictive memory when information bound up with the assembly trajectory effectively improves description or prediction after controlling for the present properties adopted. Observational results do not yet converge on a single picture. Some analyses report dependences compatible with a role for secondary halo properties, while simulations produce signals whose intensity varies with populations, selection criteria and the variables studied.
The first direct measurement carried out with data from the first data release of the DESI Bright Galaxy Survey provides a particularly instructive result. For the satellite galaxies of the sample studied, the measured parameter is Q_sat = 0.05 ± 0.14, a value compatible with zero. This result is not a mere absence of detection without theoretical consequence. The authors describe it as being in tension with many empirical models of galaxy formation. For this population and this method, the observations do not confirm the magnitude of the dependence expected by several models.
Assembly memory thus becomes an object actively put to the test. Simulations and some models produce significant secondary dependences. Several analyses find signals compatible with an influence of properties other than mass. The direct measurement of the satellites of the DESI Bright Galaxy Survey nevertheless remains compatible with the absence of a significant residual effect and thereby contradicts the expectations of many empirical models. This tension strengthens rather than weakens the scientific character of the proposal. A strong assembly memory must not be presupposed. It can fail.
If assembly history does not improve description or prediction after controlling for the relevant present variables, the claim of a residual dynamic memory fails for the population, the observables and the level of description considered. Such a result would not, however, abolish either chemical memory or galactic-archaeological reconstructibility. It would show that one particular predictive component of history is not detectable beyond the present variables adopted.
Two dimensions of regime 2 that do not depend on one another. Galactic archaeology and assembly bias are not two interchangeable examples. Gaia-Sausage-Enceladus answers mainly a retrodictive question. Can an ancient event be reconstructed from the present relations between the orbits, motions, ages and chemical compositions of stars? For the broad outlines of that accretion event, the answer is yes.
Assembly bias poses a different question. Does information associated with assembly history still improve description or prediction when certain present properties, notably halo mass, are already known? The answer depends on populations, secondary variables, methods and observables. It remains debated and may be compatible with zero in some tests.
A null result for assembly bias would not abolish reconstructible memory. Conversely, the capacity to reconstruct an ancient merger does not demonstrate that knowing that merger improves the prediction of all the galaxy's future properties. Retrodiction and the predictive value of history must be tested separately.
The second regime therefore has an internal structure. At its most directly established level, the present preserves information allowing part of the past to be reconstructed. At a more demanding level, history retains explanatory or predictive value once the variables used to describe the present state have been taken into account. The first property is clearly illustrated by chemical memory and galactic archaeology. The second is one of the stakes of research on assembly bias. Reconstructible cosmic memory does not rest on the existence of a significant assembly bias in all populations. It rests on the demonstration that certain present structures contain identifiable historical information.
Testability and descriptive relativity. A proposal about cosmic memory becomes scientifically interesting when it specifies what is supposed to persist, in which variables that persistence should be observed, and what data
could contradict the proposal. One must first define the present state used as a reference. A description limited to a halo's mass is not equivalent to a description including its concentration, its environment, its baryonic content, its internal dynamics, its stellar populations and its substructures. The richer the description, the more it can incorporate consequences of the history.
One must then define the history sought. A formation date, a number of mergers, an accretion trajectory, an episode of chemical enrichment or the origin of a stellar population are not equivalent objects. Each hypothesis requires different observables, models and criteria of success. Finally, one must specify what history adds beyond the present state. A trace has retrodictive content when it reduces the set of histories compatible with the observations. A history has residual predictive value when it improves description or prediction beyond the present variables adopted.
These two properties can be tested independently. A reconstructible memory could be put in difficulty if supposedly distinct events produced no differentiable signature, if reconstructions did not exceed what would be obtained by chance, if signatures disappeared after correction for selection biases, or if several incompatible histories remained indiscernible despite supposedly informative data. A relative predictive memory could fail if adding historical variables improved no prediction after controlling for a correctly specified present state.
Descriptive relativity does not protect the thesis from refutation. On the contrary, it obliges one to specify its domain of validity. Asserting that historical information might exist in still unknown variables does not suffice. A scientific claim must define an observable informational content, a method of comparison and conditions of failure. The proposal then becomes more limited but stronger. Some cosmic structures preserve, in certain observables and at certain resolutions, information allowing part of their history to be reconstructed. In some cases, that history can also improve description or prediction beyond a reduced macroscopic state.
Third regime: functional reuse
The third regime appears when a past trace no longer merely is present, legible or predictive for an observer. It is remobilised by the organisation and takes part causally in new cycles of maintenance, development, adaptation or reproduction. The living reaches this regime when inherited structures contribute to rebuilding, regulating or modifying the future organisation.
Genetic sequences, epigenetic states, cellular organisation, mechanisms of development and certain ecological legacies do not merely constitute traces allowing an external observer to reconstruct a history. They take part in the processes by which the organisation develops, reproduces, repairs itself and modifies its behaviour. The genome does not constitute an autonomous archive. Its sequences can be transcribed, regulated, combined and reused, but their functioning depends on an already organised cell, on molecular mechanisms, on inherited structures and on a developmental environment. The genotype–phenotype pair represents an important functional differentiation without completely separating the hereditary support from the organisation that allows it to be read.
The living does not merely carry traces. It possesses architectures able to use certain inherited structures in the production of new organised states. The difference from reconstructible cosmic memory then becomes qualitative. A galaxy may carry traces allowing observers to reconstruct its history. It does not thereby possess a device by which it would identify those traces, copy them and use them to produce a new galaxy along a hereditary cycle. A galaxy can be legible without reading itself.
The boundary between the second and third regimes does not result simply from an increase in resolution. No finer observation by itself transforms a cosmic trace into information functionally reactivated by the system that carries it. The third regime requires an additional architecture, able to preserve, recognise, transmit, transform or remobilise certain inherited structures. This boundary is qualitative in its criteria even if its historical appearance was probably gradual. Prebiotic systems may have gradually developed forms of catalysis, compartmentation, replication and transmission. There is no need to suppose an instantaneous rupture. Once the criterion is defined, the presence of a legible trace and the presence of a device that uses that trace in a new cycle nevertheless do not describe the same property.
A fundamental asymmetry between the regimes. Presenting the three regimes as homogeneous levels of a simple ladder would mask an important asymmetry. The passage from the first to the second regime depends largely on the conservation of information and on descriptive resolution. A history can become more or less legible according to the available observables, the precision of measurements, the quality of models and the time elapsed since the event. An almost entirely mixed trace can still preserve weak statistical information. A new method can render visible a hitherto inaccessible signature.
The boundary between trajectory dependence and reconstructibility is not necessarily ontological. Part of the difference comes from what one chooses to measure and from the capacity to identify the historical information contained in the present state. The passage from the second to the third regime, by contrast, introduces a different organisational property. The system no longer merely carries a trace interpretable by an external observer. It possesses internal mechanisms by which certain legacies take part in the production, maintenance or transmission of new organised states. The trace then enters a functional loop.
This asymmetry avoids two opposite errors. The first would be to reserve all memory to the living and to regard cosmic traces as mere consequences devoid of historical content. The second would be to attribute to galaxies, stars or the Universe a functional memory comparable to that of an organism merely because their present state depends on their past. A continuity exists, but it does not impose a structural isomorphism. The same general operators of partial conservation, historical dependence and transformation of constraints may appear in different architectures. Operative continuity does not abolish organisational thresholds. Where the living has developed systems of transmission allowing evolutionary accumulation, the Universe transmits chiefly landscapes of constraints, resources and potentials that each new structure must integrate or negotiate.
What persists is not necessarily useful. The comparison between cosmic and biological evolution might lead to another simplification, that systems preserve only useful traces. This formulation is too strong. The living carries legacies that are neutral, redundant, costly or that have become ill-adapted. A characteristic may persist because it remains compatible with reproduction and
development, because it is bound up with other functions, because its cost remains insufficient to bring about its elimination, or because the accessible evolutionary trajectories do not permit its immediate disappearance. Natural selection does not produce an optimal archive. It acts on historically constituted organisations constrained by their possibilities of transformation.
In a comparable way, cosmic structures do not preserve what would be useful. Some configurations persist because of their dynamic stability, their position within a gravitational environment, their history of interactions, or simply because no later event has entirely disorganised them. The Universe does not select traces according to a purpose. Structures persist because physical dynamics permit their maintenance or because their effects have not been erased.
Continuity lies in the partial and transformative character of persistence. Some traces disappear, some are incorporated, some become hard to read and some continue to modify future conditions. In the living, some additionally enter mechanisms of transmission and functional reuse.
Dissipation, entropy and non-equilibrium organisation
The comparison must not oppose dissipative cosmic structures to organisms that would preserve their order against dissipation. Cosmic structures and organisms alike appear in systems traversed by flows and far from thermodynamic equilibrium. When gas falls into a halo, it may undergo shocks, develop turbulence, radiate energy and cool. These transformations contribute to the concentration of matter and to star formation. Dark-matter halos, essentially collisionless, do not dissipate their energy in the same way. Their evolution involves notably phase mixing and processes of gravitational relaxation. Radiative dissipation must be distinguished from the mechanisms of reorganisation proper to collisionless gravitational systems.
Organisms maintain their organisation by capturing energy and matter, transforming them and then rejecting heat and degraded products into their environment. Their local order does not escape the second law of thermodynamics. It depends on continuous dissipation. The difference does not lie in the presence or absence of dissipation, but in the organisation of dissipative processes. In the living, flows are integrated into networks able to maintain certain internal conditions, produce and repair components, regulate transformations and transmit variations.
The living is not merely traversed by flows. Its organisation modifies certain relations between those flows and the internal processes on which its continuation depends. This is neither a cosmic finality nor an intention attributed to matter. It is an organisational property produced, maintained and transformed over the course of evolution. The continuity between cosmic and biological lies in the appearance of historical non-equilibrium structures. The difference appears when architectures become able to turn certain inherited traces into transmissible functional resources.
History as constraint and as a source of diversity. History does not only limit possibilities. It also produces diversity. Two halos of similar mass may follow different trajectories depending on the order of their mergers, the orientation of flows, gas inflows, their environment and the intensity of the feedbacks they
undergo. These differences can be amplified and produce galaxies whose morphologies, stellar populations or activity levels diverge.
Two organisms subject to comparable environmental constraints may likewise follow different evolutionary trajectories because of mutations, demographic events, migrations, extinctions or ecological transformations. History opens certain trajectories while closing others. This contingency does not mean that everything is possible. Physical laws, available resources, initial conditions and already constituted architectures delimit the accessible transformations. History acts within a constrained space.
Replaying an evolution from similar macroscopic conditions would not necessarily produce the same detailed succession of events. Small differences in the initial fluctuations, local interactions or merger sequences can be amplified by non-linear dynamics. The same general laws can thus produce different histories without those histories becoming arbitrary. Complexity does not result from local laws acting on matter continually reset to zero. It emerges from the interaction between relatively stable physical regularities and progressively transformed landscapes of constraint. Each structure receives a world that already has a history.
Levels of organisation that transform their own conditions. Cosmic formation may be described as a succession of levels in which each organisation modifies certain conditions accessible to the next. Initial fluctuations contribute to the formation of gravitational structures. Halos orient the accretion of baryonic matter. Gas forms stars. Stars modify the chemical composition and the energetic state of their environment. Stellar winds and explosions redistribute matter. The activity of black holes can modify the evolution of gas at the scale of galaxies and clusters.
Each level appears within an already transformed environment and then contributes to transforming future environments. Structures produce constraints that subsequently become the conditions of new structures. This dynamic presupposes no cosmic programme. It does not mean that the Universe necessarily tends toward life, consciousness or knowledge. Trajectories remain contingent and the majority of environments probably never produce living systems.
It does show, however, that the emergence of a level cannot always be understood by examining only the local laws present at the moment of its appearance. One must also take into account the inherited conditions that make certain transformations accessible and others improbable. The living appears in a chemically enriched Universe, structured by generations of stars and able to produce certain durable planetary environments. These conditions do not suffice to explain the appearance of life, but they constitute part of its material history. Cosmic memory does not designate a faculty added to matter after the fact. It designates the persistence of certain transformations within the conditions offered to subsequent organisations.
A thesis independent of any single cosmological model
This article uses mainly ΛCDM because most simulations of structure
formation and analyses of the relation between galaxies and halos are today developed within that framework. The central proposal is nevertheless more general. Some present structures may contain measurable information about their history and some dynamics may depend on constraints produced by the earlier formation of structures.
The precise mechanisms, their importance and their interpretation may vary between cosmological models. Approaches founded on Buchert averaging grant an explicit role to the effects of inhomogeneities on average evolution. The averaged equations introduce a kinematical backreaction and relate the evolution of the average curvature to the development of structures. Within this framework, average evolution may depend on how the contrasts of density and expansion were constituted over time.
This relation may be interpreted cautiously as a form of first-regime trajectory dependence at a more global scale. The average state would then not be entirely independent of the history of structure formation. This interpretation nevertheless demonstrates neither a reconstructible memory of space-time nor a functional memory. It indicates only that, in such frameworks, the history of structuring may take part in the average dynamics rather than remain a local detail without feedback.
Wiltshire's timescape model is a particular realisation developed from Buchert's formalism. It attributes a role to the differences of evolution between dense regions and voids as well as to the relative calibration of clocks and distances in an inhomogeneous Universe. It proposes observables allowing it to be distinguished from homogeneous cosmologies containing a cosmological constant. Cosmic memory does not, however, constitute an automatic argument in favour of ΛCDM, of Buchert or of timescape. It designates a set of empirical properties whose form, intensity and legibility can be compared between several cosmological descriptions.
In ΛCDM, memory is mainly studied in assembly histories, matter distributions, stellar populations and chemical signatures. In inhomogeneous cosmologies, certain forms of historical dependence could also intervene in the average evolution. Each framework must specify its mechanisms and produce its own criteria of testing.
An increasingly legible memory
Contemporary instruments are rapidly increasing the quantity and precision of accessible traces. Gaia measures the positions, motions and various properties of a very large number of stars in the Milky Way. These data make it possible progressively to reconstruct the dynamics of the disc, the halo and several ancient populations. The mission has profoundly widened the possibility of relating present stellar properties to the assembly history of the Galaxy.
Euclid maps the distribution and morphology of millions of galaxies. Its first quick data release, published on 19 March 2025, provided a glimpse of its deep fields and about twenty-six million detections. That first release was not yet a complete cosmological publication, but it made vast sets of galactic structures and their large-scale organisation accessible.
DESI is building a three-dimensional representation of the distribution of galaxies and quasars over vast volumes. Its data make it possible not only to study the history of expansion, but also to test more finely the relations between galaxies, halos and environment. The direct measurement of satellite assembly bias shows that increasing precision does not necessarily confirm theoretical expectations and may also put them in difficulty. JWST and ALMA observe the first phases of galaxy formation and enrichment. The detections made in JADES-GS-z14-0 show that a significant chemical evolution had already begun about three hundred million years after the Big Bang.
These instruments do not merely make distant objects visible. They give access to different dimensions of the material history of the cosmos, from the geometry of the cosmic web to the motions of galaxies, matter distributions, stellar populations and chemical abundances. The increase in this legibility does not guarantee that all traces will survive, nor that a single history will be reconstructible. Some information is lost. Some transformations produce ambiguous signatures. Several trajectories can lead to similar states. Reconstruction remains partial, probabilistic and model-dependent. An incomplete memory is nevertheless not an absence of memory.
Conclusion
The Universe does not work like a single shell preserving each of its stages in intact layers. Its history is made of accretion, merger, dispersion, mixing and reorganisation. Old structures may disappear and part of the information they carried may become inaccessible. Yet new structures do not form in a virgin space. They appear within a real already differentiated by its history. Matter distributions orient future flows. Stars transform the composition of the medium from which the following generations will form. Absorbed galaxies sometimes leave populations whose properties allow part of their origin to be reconstructed. Present structures thus modify the conditions under which new structures can appear.
This historical continuity can be distinguished into three regimes. In the first, the past takes part in producing the present state. This is trajectory dependence. In the second, certain transformations leave information sufficiently organised to permit a partial reconstruction of the history. This regime has two distinct dimensions. The first is retrodictive, since the present allows part of the past to be inferred. Chemical memory and galactic archaeology provide solid examples. The second is predictive, since history still supplies information beyond the variables chosen to describe the present state. Research on assembly bias tests this more demanding claim, whose results remain divided and may be compatible with zero for some populations. In the third regime, certain organisations reuse inherited traces within new cycles of maintenance, development and transmission. Memory then becomes functional.
The boundary between trajectory dependence and reconstructibility is gradual and relative to the level of description. An improvement in observations, a change of variables or a new method can render visible historical information previously inaccessible. The boundary between reconstructibility and functional reuse introduces an organisational difference. A legible trace does not automatically become a resource used by the system that carries it. The cosmic clearly reaches historical dependence and, in many cases, reconstructibility. The living adds architectures of transmission, reading and reactivation that permit evolutionary accumulation.
This distinction preserves both the continuity and the difference. Cosmic and biological systems are not structurally identical, but their present does not result from local laws acting on a blank slate. It emerges from the interaction between those laws and an already transformed landscape of constraints. Matter does not preserve a faithful copy of everything that happened to it. Some transformations leave only a dynamic dependence. Others produce still legible traces. Some retain explanatory or predictive value beyond a simplified present description. In the living, some traces become transmissible and reusable resources.
Cosmic memory is less recorded than incorporated. It resides in the geometry, the composition and the dynamics of structures. Its scientific reach begins where that incorporation produces historical information that is identifiable, measurable and liable to be falsified. The emergence of complexity does not result solely from local physical laws applied to matter without a history. It also depends on the accumulation, transformation and reorganisation of inherited constraints. Laws delimit the possible transformations, while history contributes to determining which paths actually become accessible.