The approximate homogeneity of spatial sections of the Universe is well supported observationally, but the inhomogeneity of the spatial sections is even better supported. Here, we consider the implications of inhomogeneity in dust models for the connectedness of spatial sections at early times. We consider a non-global Lemaitre-Tolman-Bondi (LTB) model designed to match observations, a more general, heuristic model motivated by the former, and two specific, global LTB models. We propose that the generic class of solutions of the Einstein equations projected back in time from the spatial section at the present epoch includes subclasses in which the spatial section evolves (with increasing time) smoothly (i) from being disconnected to being connected, or (ii) from being simply connected to being multiply connected, where the coordinate system is comoving and synchronous. We show that (i) and (ii) each contain at least one exact solution. These subclasses exist because the Einstein equations allow non-simultaneous big bang times. The two types of topology evolution occur over time slices that include significantly post-quantum epochs if the bang time varies by much more than a Planck time. In this sense, it is possible for cosmic topology evolution to be "mostly" classical.

Compact, flat Friedmann-Lemaitre-Robertson-Walker (FLRW) models have recently regained interest as a good fit to the observed cosmic microwave background temperature fluctuations. However, it is generally thought that a globally, exactly-flat FLRW model is theoretically improbable. Here, in order to obtain a probability space on the set F of compact, comoving, 3-spatial sections of FLRW models, a physically motivated hypothesis is proposed, using the density parameter Omega as a derived rather than fundamental parameter. We assume that the processes that select the 3-manifold also select a global mass-energy and a Hubble parameter. The inferred range in Omega consists of a single real value for any 3-manifold. Thus, the obvious measure over F is the discrete measure. Hence, if the global mass-energy and Hubble parameter are a function of 3-manifold choice among compact FLRW models, then probability spaces parametrised by Omega do not, in general, give a zero probability of a flat model. Alternatively, parametrisation by the injectivity radius r_inj ("size") suggests the Lebesgue measure. In this case, the probability space over the injectivity radius implies that flat models occur almost surely (a.s.), in the sense of probability theory, and non-flat models a.s. do not occur.

Observational evidence of dark energy that makes the Universe nearly flat at the present epoch is very strong. We study the link between spatial continuity and dark energy. We assume that comoving space is a compact 3-manifold of constant curvature, described by a homogeneous Friedman-Lemaitre-Robertson-Walker metric. We assume that spatial continuity cannot be violated, i.e. that the global topology of the comoving section of the Universe cannot change during post-quantum epochs. We find that if the Universe was flat and compact during early epochs, then the presently low values of the radiation and matter densities imply that dark energy was created as a spatial continuity effect. Moreover, if the Universe is compact, then Omega_tot=1 is dynamically stable, where Omega_tot is the total density parameter in units of the critical density. Dark energy was observationally detected as a geometrical phenomenon. It is difficult to imagine a simpler explanation for dark energy than spatial continuity, finiteness and homogeneity.

Quel est, ou pourrait \^etre, la topologie globale de la partie spatiale de l'Univers ? L'Univers entier (pr\'ecis\'ement, l'hypersurface spatiale de celui-ci) est-il observable ? Les math\'ematiciens, les physiciens et les cosmologistes observationnels ont des approches diff\'erentes pour aborder ces questions qui restent ouvertes. What is, or could be, the global topology of spatial sections of the Universe? Is the entire Universe (spatial hypersurface thereof) observable ? Mathematicians, physicists and observational cosmologists have different strategies to approaching these questions which are not yet fully answered.

The space-like hypersurface of the Universe at the present cosmological time is a three-dimensional manifold. A non-trivial global topology of this space-like hypersurface would imply that the apparently observable universe (the sphere of particle horizon radius) could contain several images of the single, physical Universe. Recent three-dimensional techniques for constraining and/or detecting this topology are reviewed. Initial applications of these techniques using X-ray bright clusters of galaxies and quasars imply (weak) candidates for a non-trivial topology.