For the past twenty years, Mathematical Finance has grown from the perfect fit between martingale methods and models of frictionless markets. But in last two years, the limits of this theory have become painfully clear, with the widespread failure of the valuation and risk control systems in the financial industry.This proposal lays the groundwork for a new generation of models, which include nonlinear frictions such as transaction costs and liquidity as essential elements, not as extra features. This endeavor entails developing new notions of nonlinear stochastic integrals, and requires a theory that looks beyond the established setting of semimartingales. To become useful, this theory will need tools to solve related optimization problems, either explicitly, or with asymptotic methods. Convex duality and control theory will help develop such tools, together with partial differential equations techniques.The proposed research aims at (i) understanding the natural setting of frictions models from well-posedness principles, (ii) developing a consistent integration theory, and (iii) investigating implications for optimization problems. These steps are central to nurture a new class of financial models, which can eventually remedy the pitfalls of the current ones.