In this thesis, we present ab initio studies on atomic subsystems in contact with extended environments. Among a large number of systems which can be treated in the subsystem-environment framework, here we consider the two typical cases of mechanical and of electronic interaction with the environment. Specifically, we consider atomic deformations in defect-containing crystals and photo-excitation processes of solvated molecules. In the former case, the subsystem is a group of atoms involving particular microscopic displacements, which turn out to be key ingredients in understanding structural phase transitions and glass-like low-energy excitations reported to exist in certain apparent crystalline systems. In this case, the crystal surrounding the displaced atoms makes up the the environment, which couples via effective elastic strain fields to the subsystem. The subsystem itself, then, we find can be described quite accurately by a simple model Hamiltonian which captures all of the key behaviors found in the ab initio calculations. This approach allows us to efficiently handle complicated environments, such as stacking faults and randomly distributed dopants, which would otherwise be impractical to study directly ab initio. This approach then allows us to successfully predict complex glass-like system behaviors in good agreement with experimental observations. In the latter case of photo-excitations, we focus on non-equilibrium states of molecules in the quasiparticle framework based on the Green's function approach within the "GW" approximation. The environment in this case is made up of the surrounding polarizable liquid molecules, which we describe in this work via one of two different approaches: the classical model of a polarizable continuum (PCM method) and the chemistry-inspired cluster expansion of microscopic polarization (a new method which we introduce here.) We show that an appropriate combination of the two embedding approaches successfully captures the environmental effect. We thus bring significant improvement to the standard PCM model in computing solvation shifts of the ionization potential, while requiring much less computational power than would direct computation with the entire environment included explicitly. In addition, our approach allows us to develop a systematic improvement over state-of-the-art PCM methods, which are highly efficient computationally. Through construction of a microscopic dielectric model for the environment based on microscopic molecular polarizations, we are able to derive a non-local dielectric description which includes short-wavelength features missing in the standard treatment of the PCM method.