The popularity of the concept of digital twin is sensitively growing across different domains and a variety of definitions have been adopted across science and engineering communities. This minisymposium builds up on the definition of digital twins as virtual models that progressively adapt and specialize by learning from data from their real (natural or artificial) counterparts [1,2] and aims at offering a forum to explore the pivotal role that this specific class of models can play to enable a sustainable development. In particular, a twofold relevance of digital twins to UN SDGs [3] will be considered: (1) enabling better informed and timely decisions along a sustainability-driven product life cycle; (2) enabling data-efficient learning as opposed to energy-intensive state of the art methods. Accordingly, contributions are invited that address the usefulness of digital twins and meet the purpose they have been developed for; that is their ability to provide timely and reliable predictions in support of design, manufacturing, maintenance, operational or decommissioning decision problems. In addition, contributions are solicited on mathematical approaches and computational strategies that enable digital twins to efficiently assimilate in small data settings and contain the demand of computing and energy resources; methodological advances across data assimilation, uncertainty quantification, model order reduction and scientific machine learning are most welcome.