In collaboration with UC Berkeley and Tokyo University, the seismic source is studied through non-equilibrium thermodynamics and nonlinear mechanics, disposing of the paradigm of Elastic Rebound. The latter, which is based on linear elasticity and considers only the mechanical terms, is notoriously capable of effective representations of the wavefield, but is ineffective in describing the evolution of the process, which is reduced to a fracture + stick slip on elementary geometry. This leads not only to the impotence of forecasting future events, but also to several paradoxes, the first of which relates to the amount of heat produced by the earthquake sliding. The new approach addresses the problem of seismic source ab initio, by combining the thermodynamic and mechanical problems, and taking into account the presence of fluids and of a non-Euclidean geometry. This resolves the paradoxes, reconciling theory with experiment, with earthquakes occurring in clusters which are self-similar in both space and time. The new approach is also effective in describing the seismicity induced by fluid injection, reconciling the theory with experimental evidence, that sees earthquakes induced even at considerable distances and time delays from the injection, with the fluid diffusing in the solid matrix as pressure waves. A similar effect is true for the occurrence of aftershocks at spatial and temporal distances which are several orders of magnitude larger than expected from the classical approach.