We discuss two procedures to obtain empirical potentials from ab initio trajectories. The first methodconsists in adjusting the parameters of an empirical pair potential so that the radial distribution functionsextracted from classical simulations using this potential match the ones extracted from the ab initio sim-ulations. As a case study, we consider the example of amorphous silica, a material that is highly relevantin the field of glass science as well as in geology. With our approach we are able to obtain an empiricalpotential that gives a better description with respect to structural and thermodynamic properties thanthe potential proposed by van Beest, Kramer, and van Santen, and that has been very frequently usedas a model for amorphous silica. The second method is the so-called ‘‘force matching” approach proposedby Ercolessi and Adams to obtain an empirical potential. We demonstrate that for the case of silica thismethod does not yield a reliable potential and discuss the likely origin for this failure.

The vibrational properties of silica glass have been intensively studied experimentally andtheoretically during the last four decades. However there are few theoretical studies of theevolution of the vibrational properties for more complex silicate glasses (binary, ternary etc...).In the first part of this talk, we will present the structural and vibrational properties a sodium silicateglass of composition 4Na 2 O-6SiO 2 (NS1.5). The study has been carried out using first principlescalculations within the density functional theory framework as implemented in the VASP code. Wehave studied the structural and vibrational properties of this glass, and we have identified thecontributions of the various species to the vibrational density of states (VDOS) as well as to the IRspectra. In particular we will discuss how the presences of the sodium atoms affect the relevantvibrational parameters: positions, shapes and intensities of the main peaks of the VDOS and IRspectra.In the second part, we will discuss the properties of several complex borosilicate glasses.Computing physical properties of complex glass compositions via numerical simulations oftensuffers of inaccurate predictions when using classical potentials, or it is excessively timeconsuming simulations when glass structures are obtained using ab-initio methods. The study hasbeen carried out by combining classical and ab initio molecular dynamics (MD) methods and byvarying the concentrations of SiO 2 , B 2 O 3 , Na 2 O, CaO, Al 2 O 3 and MgO. The classical MD includedliquid equilibration, quenching and preliminary relaxation, while ab initio method was used tofurther relax the samples. Two sets of effective potentials have been used for the classicalsimulations. This strategy allows of course a significant CPU time saving.We have compared the structural properties of the simulated samples and have found thatclassical MD simulations cannot provide reliable glass models for complex borosilicates (at leastwith the potentials we have chosed), especially for the peculiar aspect of Boron coordination. Theab initio simulations have corrected the unreliable classical simulated structures and providedresults in good agreement with previous studies of borosilicate glasses. We have also noticed thatthe presence of Al and Mg has influenced the fraction of 3-fold coordinated Boron, as expectedfrom experimental data. The vibrational properties have been equally studied within the ab initioapproach, and the contributions of the various species have been identified. We have found that 3-and 4-fold coordinated borons give rise to distinguished spectral features.