We introduce a new model for the spectral energy distribution of galaxies, GRASIL-3D, which includes a careful modelling of the dust component of the interstellar medium. GRASIL-3D is an entirely new model based on the formalism of an existing and widely applied spectrophotometric model, GRASIL, but specifically designed to be interfaced with galaxies with any arbitrarily given geometry, such as galaxies calculated by theoretical hydrodynamical galaxy formation codes. GRASIL-3D is designed to separately treat radiative transfer in molecular clouds and in the diffuse cirrus component. The code has a general applicability to the outputs of simulated galaxies, either from Lagrangian or Eulerian hydrodynamic codes. As an application, the new model has been interfaced to the P-DEVA and GASOLINE smoothed-particle hydrodynamic codes, and has been used to calculate the spectral energy distribution for a variety of simulated galaxies from UV to sub-millimeter wavelengths, whose comparison with observational data gives encouraging results. In addition, GRASIL-3D allows 2D images of such galaxies to be obtained, at several angles and in different bands.

The spectral energy distribution of galaxies is a complex function of the star formation history and geometrical arrangement of stars and gas in galaxies. The computation of the radiative transfer of stellar radiation through the dust distribution is time-consuming. This aspect becomes unacceptable in particular when dealing with the predictions by semi-analytical galaxy formation models populating cosmological volumes, to be then compared with multi-wavelength surveys. Mainly for this aim, we have implemented an artificial neural network algorithm into the spectro-photometric and radiative transfer code GRASIL in order to compute the spectral energy distribution of galaxies in a short computing time. This allows to avoid the adoption of empirical templates that may have nothing to do with the mock galaxies output by models. The ANN has been implemented to compute the dust emission spectrum (the bottleneck of the computation), and separately for the star-forming molecular clouds and the diffuse dust (due to their different properties and dependencies). We have defined the input neurons effectively determining their emission, which means this implementation has a general applicability and is not linked to a particular galaxy formation model. We have trained the net for the disc and spherical geometries, and tested its performance to reproduce the SED of disc and starburst galaxies, as well as for a semi-analytical model for spheroidal galaxies. We have checked that for this model both the SEDs and the galaxy counts in the Herschel bands obtained with the ANN approximation are almost superimposed to the same quantities obtained with the full GRASIL. We conclude that this method appears robust and advantageous, and will present the application to a more complex SAM in another paper.

We present photometric evolution models of galaxies, in which, in addition to the stellar component, the effects of an evolving dusty interstellar medium have been included with particular care. Starting from the work of Calura, Pipino & Matteucci (2008), in which chemical evolution models have been used to study the evolution of both the gas and dust components of the interstellar medium in the solar neighbourhood, elliptical and irregular galaxies, it has been possible to combine these models with a spectrophotometric stellar code that includes dust reprocessing (GRASIL) (Silva et al. 1998) to analyse the evolution of the spectral energy distributions (SED) of these galaxies. We test our models against observed SEDs both in the local universe and at high redshift and use them to predict how the percentage of reprocessed starlight evolves for each type of galaxy. The importance of following the dust evolution is investigated by comparing our results with those obtained by adopting simple assumptions to treat this component.