This chapter gives a concise summary of various ideas, conceptual arguments, machine-learning algorithms, theory-basedlearning, and implementation procedures for the incorporation of urban dynamics and complexity into digital twins of citiesfor the effective and efficient modeling, simulation, prediction, and management of our ever-evolving cities. It alsoproposes directions for future research along the line of elaborating and strengthening the theoretical underpinnings of theapproaches in the proposed framework and experimentally verifying and validating them in synthetic simulations and reallifeapplications; further developing methods to learn the low-dimensional manifold of the extremely high-dimensional datato achieve parsimony beneficial to digital-twin construction and learning; facilitating our multiple stakeholder decisionmaking; and designing operational systems for the simulation, prediction, and management of large-scale urban systemswith continuous knowledge and data flows among the physical city entities, the corresponding digital twins and the datahubs in a synchronised and timely manner. It is envisioned that it will advance a new frontier of modeling geographicalsystems in general and urban systems, together with their digital twins, in particular. It will also break new grounds ininterdisciplinary research involving academia, industry, government, and practicing professionals on human and physical systems that are ever evolving in space and time.