The evolution of the perturbations in the energy density and the particle number density in a flat Friedmann-Lemaitre-Robertson-Walker universe in the radiation-dominated era and in the epoch after decoupling of matter and radiation is studied. For large-scale perturbations the outcome is in accordance with treatments in the literature. For small-scale perturbations the differences are conspicuous. Firstly, in the radiation-dominated era small-scale perturbations grew proportional to the square root of time. Secondly, perturbations in the Cold Dark Matter particle number density were, due to gravitation, coupled to perturbations in the total energy density. This implies that structure formation could have begun successfully only after decoupling of matter and radiation. Finally, after decoupling density perturbations evolved diabatically, i.e., they exchanged heat with their environment. This heat exchange may have enhanced the growth rate of their mass sufficiently to explain structure formation in the early universe, a phenomenon which cannot be understood from adiabatic density perturbations.

It is shown that the decomposition theorems of York, Stewart and Walker for symmetric spatial second-rank tensors, such as the perturbed metric tensor and perturbed Ricci tensor, and the spatial fluid velocity vector imply that, for open, flat or closed Friedmann-Lemaitre-Robertson-Walker universes, there are exactly two, unique, independent gauge-invariant quantities which describe the perturbations to the energy density and particle number density. Using these two new quantities, evolution equations for cosmological density perturbations, adapted to non-barotropic equations of state for the pressure, are derived. The new definitions for the perturbations to the energy density and particle number density allow for an exact non-relativistic limit with a time-independent Newtonian potential. It is shown that density perturbations evolve adiabatically if and only if the particle number density does not contribute to the pressure.

It is shown that a first-order cosmological perturbation theory for the open, flat and closed Friedmann-Lema\^itre-Robertson-Walker universes admits one, and only one, gauge-invariant variable which describes the perturbation to the energy density and which becomes equal to the usual Newtonian energy density in the non-relativistic limit. The same holds true for the perturbation to the particle number density. Using these two new variables, a new manifestly gauge-invariant cosmological perturbation theory has been developed. Density perturbations evolve diabatically. Perturbations in the total energy density are gravitationally coupled to perturbations in the particle number density, irrespective of the nature of the particles. There is, in first-order, no back-reaction of perturbations to the global expansion of the universe. Small-scale perturbations in the radiation-dominated era oscillate with an increasing amplitude, whereas in older, less precise treatments, oscillating perturbations are found with a decreasing amplitude. This is a completely new and, obviously, important result, since it makes it possible to explain and understand the formation of massive stars after decoupling of matter and radiation.

It is shown that a first-order relativistic perturbation theory for the open, flat or closed Friedmann-Lemaitre-Robertson-Walker universe admits one, and only one, gauge-invariant quantity which describes the perturbation to the energy density and which becomes equal to the usual Newtonian energy density in the non-relativistic limit. The same holds true for the perturbation to the particle number density. These facts exclude all definitions of gauge-invariant quantities used to describe density perturbations in former theories. Using these two new quantities, a manifestly covariant and gauge-invariant cosmological perturbation theory, adapted to non-barotropic equations of state for the pressure, has been developed. The new theory is valid for all scales since metric gradients do not occur in the final evolution equations. The new theory has an exact non-relativistic limit with a time-independent Newtonian potential. The usual Newtonian perturbation theory is inadequate to study the evolution of density perturbations. In the radiation-dominated era, perturbations in the particle number density are gravitationally coupled to perturbations in the total energy density, irrespective of the nature of the particles. This implies that structure formation can commence only after decoupling of matter and radiation. After decoupling of matter and radiation density perturbations evolve diabatically, i.e., they exchange heat with their environment. This heat loss of a perturbation may enhance the growth rate of its mass sufficiently to explain stellar formation in the early universe, a phenomenon not understood, as yet, without the additional assumption of the existence of Cold Dark Matter. This theoretical observation is the main result of this article.

It is shown that a first-order cosmological perturbation theory for Friedmann-Lemaitre-Robertson-Walker universes admits one and only one gauge-invariant variable which describes the perturbation to the energy density and which becomes equal to the usual energy density of the Newtonian theory of gravity in the limit that all particle velocities are negligible with respect to the speed of light. The same holds true for the perturbation to the particle number density. A cosmological perturbation theory based on these particular gauge-invariant quantities is more precise than any earlier first-order perturbation theory. In particular, it explains star formation in a satisfactory way, even in the absence of cold dark matter. In a baryon-only universe, the earliest stars, the so-called Population III stars, are found to have masses between 400 and 100,000 solar masses with a peak around 3400 solar masses. If cold dark matter, with particle mass 10 times heavier than the proton mass, is present then the star masses are between 16 and 4000 solar masses with a peak around 140 solar masses. They come into existence between 100 Myr and 1000 Myr. At much later times, star formation is possible only in high density regions, for example within galaxies. Late time stars may have much smaller masses than early stars. The smallest stars that can be formed have masses of 0.2--0.8 solar mass, depending on the initial internal relative pressure perturbation. It is demonstrated that the Newtonian theory of gravity cannot be used to study the evolution of cosmological density perturbations.

Contrarily to general believe, a first-order cosmological perturbation theory based on Einstein's General Theory of Relativity explains the formation of massive primeval stars in a flat Friedmann-Lemaitre-Robertson-Walker universe after decoupling of matter and radiation, whether or not Cold Dark Matter is present. The growth rate of a density perturbation depends on the heat loss of a perturbation during the contraction, but is independent of the particle mass. The relativistic Jeans mass does depend on the particle mass. If the Cold Dark Matter particle mass is equal to the proton mass, then the relativistic Jeans mass is equal to 3500 solar masses, whereas the classical Jeans mass is a factor 145 larger.

We propose a way to construct manifestly gauge independent quantities out of the gauge dependent quantities occurring in the linearized Einstein equations. Thereupon, we show that these gauge-invariant combinations can be identified with measurable perturbations to the particle and energy densities. In the radiation-dominated era we find, for small-scale perturbations, acoustic waves with an increasing amplitude, while standard treatments predict acoustic waves with a decaying amplitude. For large-scale perturbations we find exactly the same growth rates as in the standard literature. When considering the non-relativistic limit of the linearized Einstein equations we find the Poisson equation. It is shown, using the linearized Einstein equations, that the usual Newtonian treatment of density perturbations does not describe the evolution of density perturbations.