In the context of the climate crisis, fuel sprays are of great societal importance because they directly impact energy consumption and pollutant emissions in liquid-fueled combustion systems. More broadly, liquid sprays appear in a wide range of applications such as powder production for additive manufacturing, drug encapsulation, and coatings for materials. Because of their importance and ubiquity, there is a strong desire to better model the spray atomization process and to develop optimization and control strategies to improve efficiency and to aid in the design of spray systems. However, computational modeling of atomization presents enormous challenges, in part because the extreme complexity of the associated flow field and because of the wide range of length and time scales involved. In the first part of this work, I present a multi-scale simulation strategy that models the full spray atomization region from first principles. In contrast to prior work, this approach simulates the atomization process end-to-end, i.e., from the inlet of the injector to the spray dispersion region downstream. To that end, multiple simulation domains are coupled together to address the wide range of length scales that need to be captured. Another significant departure from standard approaches is that multiple sub-grid scale models are introduced to account for unresolved processes, most importantly a thin structure break-up model to explicitly model topology change. Simulations of a canonical two-fluid atomizer are quantitatively validated against experiments at identical operating conditions, including drop size statistics. As the computational prediction of liquid-gas flows remains arduous, it is no surprise that numerical frameworks for optimal control of liquid-gas flows have not received much attention from researchers yet. In the second part of this work, I present one of the very first computational adjoint frameworks for the optimization of liquid-gas flows using a sharp interface model. This approach is verified by showing low gradient errors for a range of test cases. It is then used to maximize the growth of a temporally evolving liquid-gas mixing layer, which demonstrates the potential of the proposed method for optimizing spray formation.