We present an enhanced reverse Monte Carlo approach that includes fitting to NMR data in the form of chemical shifts in addition to the usually used scattering data. Furthermore, the internal energy is accounted for in the cost function to prevent unphysical structures. This approach was applied to generate structural models of amorphous Si₃B₃N₇, which is a prototype of a new class of high performance ceramics exhibiting interesting features like high thermal and mechanical stability.

We fitted our structural models in direct space to radial distribution functions from x-ray, neutron and electron scattering experiments, and to the ¹⁵N NMR spectrum of the ceramic. This spectrum could not be interpreted before since it exhibits a broad structureless signal which is a superposition of peaks related to different chemical environments NB_xSi_3−x (x = 0–3) whose chemical shifts were only partly known experimentally. Therefore we based the calculation of NMR data in the reverse Monte Carlo optimizations on previous theoretical work that was done in our group. All generated models reproduce the experimental radial distribution functions very well. This good agreement does not deteriorate when the NMR data are taken into account. Fitting the models to ¹⁵N NMR chemical shifts in addition to scattering data results in structural changes that not only improve the agreement with the experimental magic-angle spinning (MAS) NMR spectrum but yield also significantly better second-nearest neighbour coordination statistics.