We investigate $f(R,T)$ gravity models ($R$ is the curvature scalar and $T$ is the trace of the stress-energy tensor of ordinary matter) that are able to reproduce the four known types of future finite-time singularities. We choose a suitable expression for the Hubble parameter in order to realise the cosmic acceleration and we introduce two parameters, $\alpha$ and $H_s$, which characterise each type of singularity. We address conformal anomaly and we observe that it cannot remove the sudden singularity or the type IV one, but, for some values of $\alpha$, the big rip and the type III singularity may be avoided. We also find that, even without taking into account conformal anomaly, the big rip and the type III singularity may be removed thanks to the presence of the $T$ contribution of the $f(R,T)$ theory.

Rastall's theory is a modification of General Relativity, based on the non-conservation of the stress-energy tensor. The latter is encoded in a parameter $\gamma$ such that $\gamma = 1$ restores the usual $\nabla_\nu T^{\mu\nu} = 0$ law. We test Rastall's theory in cosmology, on a flat Robertson-Walker metric, investigating a two-fluid model and using the type Ia supernovae Constitution dataset. One of the fluids is pressureless and obeys the usual conservation law, whereas the other is described by an equation of state $p_x = w_x\rho_x$, with $w_x$ constant. The Bayesian analysis of the Constitution set does not strictly constrain the parameter $\gamma$ and prefers values of $w_x$ close to -1. We then address the evolution of small perturbations and show that they are dramatically unstable if $w_x \neq -1$ and $\gamma \neq 1$, i.e. General Relativity is the favored configuration. The only alternative is $w_x = -1$, for which the dynamics becomes independent from $\gamma$.

We review the difficulties of the generalized Chaplygin gas model to fit observational data, due to the tension between background and perturbative tests. We argue that such issues may be circumvented by means of a self-interacting scalar field representation of the model. However, this proposal seems to be successful only if the self-interacting scalar field has a non-canonical form. The latter can be implemented in Rastall's theory of gravity, which is based on a modification of the usual matter conservation law. We show that, besides its application to the generalized Chaplygin gas model, other cosmological models based on Rastall's theory have many interesting and unexpected new features.

Rastall's theory is based on the non-conservation of the energy-momentum tensor. We show that, in this theory, if we introduce a two-fluid model, one component representing vacuum energy whereas the other pressureless matter (e.g. baryons plus cold dark matter), the cosmological scenario is the same as for the \Lambda CDM model, both at background and linear perturbative levels, except for one aspect: now dark energy may cluster. We speculate that this can lead to a possibility of distinguishing the models at the non-linear perturbative level.

We find exact power-law solutions for scalar-tensor theories and clarify the conditions under which they can account for an accelerated expansion of the Universe. These solutions have the property that the signs of both the Hubble rate and the deceleration parameter in the Jordan frame may be different from the signs of their Einstein-frame counterparts. For special parameter combinations we identify these solutions with asymptotic attractors that have been obtained in the literature through dynamical-system analysis. We establish an effective general-relativistic description for which the geometrical equivalent of dark energy is associated with a time dependent equation of state. The present value of the latter is consistent with the observed cosmological ``constant". We demonstrate that this type of power-law solutions for accelerated expansion cannot be realized in f(R) theories.

The Rastall's theory is a modification of the General Relativity theory leading to a different expression for the conservation law in the matter sector compared with the usual one. It has been argued recently that such a theory may have applications to the dark energy problem, since a pressureless fluid may lead to an accelerated universe. In the present work we confront the Rastall's theory with the power spectrum data. The results indicate a configuration that essentially reduces the Rastall's theory to General Relativity, unless the non-usual conservation law refers to a scalar field, situation where other configurations are eventually possible.

The dynamics of TeVeS in a homogeneous and isotropic universe is shown to be equivalent to the dynamics of an interacting two-component system, consisting of a scalar field and a "fluid", related to the matter part, with explicitly given coupling term. Scaling solutions (solutions with a constant ratio of the energy densities of both components) in the "Einstein frame" are found which are exponentially expanding or contracting with no remaining freedom for Bekenstein's $F$ function. In the "physical frame" these solutions are of the power-law type. An equivalent General Relativity (GR) picture of the dynamics suggests that it is the scalar field which plays the role of dark matter, while the "matter" has to mimic (phantom-type) dark energy.

The Chaplygin gas model, characterized by an equation of state of the type $p = - \frac{A}{\rho}$ emerges naturally from the Nambu-Goto action of string theory. This fluid representation can be recast under the form of a tachyonic field given by a Born-Infeld type Lagrangian. At the same time, the Chaplygin gas equation of state can be obtained from a self-interacting scalar field. We show that, from the point of view of the supernova type Ia data, the three representations (fluid, tachyonic, scalar field) lead to the same results. However, concerning the matter power spectra, while the fluid and tachyonic descriptions lead to exactly the same results, the self-interacting scalar field representation implies different statistical estimations for the parameters. In particular, the estimation for the dark matter density parameter in the fluid representation favors a universe dominated almost completely by dark matter, while in the self-interacting scalar field representation the prediction is very closed to that obtained in the $\Lambda$CDM model.