The question of 'predictability' in multi-scale materials simulations is a very important one, since we want the results to be reliable, qualitatively or quantitatively. The frist step in achieveing this largely depends upon having a source of accurate quantum chemical forces that underlie the classical dynamics. However, unlike small molecules, where highly sophisticated and predictive methods like coupled-cluster theory can be applied, for any multi-scale simulation hundreds to thousands of atoms need to be described by the quantum chemical methods with an efficientcy that permits tying to dynamics. To solve this problem we have embarked upon the concept of a 'transfer Hamiltonian' which is formally defined by coupled-cluster theory whose eigenfunction is a single determinant, yet whose energy and forces are exact. Furthermore, it also permits the treatment of different electronic or ionized states permitting optical properties and state specificity to be described as part of the simulation. This transfer Hamiltonian can be evaluated purely from first principles, or from determining parameters that define the Hamiltonian for the particular phenomena of interest. We will illustrate this approach in comparison with others like density functional theory in problems involving the fracture of silica including the presence of water.