A digital twin is a virtual representation of a physical object, designed to exactly mimic its inspiration — not as a simple model, but as a complex system that uses real-time data to change in all the same ways as its real-world counterpart. The potential applications are far-reaching. In health care, the digital twin of a person might predict how their body would respond to a new drug or therapy. In aerospace engineering, the digital twin of an airplane could predict its responses to atmospheric changes.
Digital twins have drawn the attention of researchers across many disciplines, says Yuanzhao Zhang, a Complexity Postdoctoral Fellow at SFI. But some challenges remain before the technology can be reliably deployed to challenging domains such as healthcare, he says.
 |
| Participants in SFI's August 2025 Digital Twins working group. [From top: Declan Norton, Pepper Huang, Anthony Gruber, Serkan Gugercin, Karen Willcox, and Yuanzhao Zhang (not pictured: Luwen Huang, Michael Mahoney, Michael Kapteyn] (image: Yuanzhao Zhang) |
To address those challenges, Zhang and engineer Karen Willcox, an SFI External Faculty member and Director of the Oden Institute for Computational Engineering and Sciences at the University of Texas at Austin, invited a small group of mathematicians, computer scientists, physicists, and others to a working group held at SFI on August 12 and 13, 2025. The working group focused on ways to leverage recent work on dynamical systems, network theory, and information theory to tackle some of the largest challenges in digital twins.
One of the biggest challenges, says Zhang, is managing an enormous amount of information. To accurately represent a dynamic system, a digital twin must quickly process a stream of incoming data, which incurs a hefty computational cost. “What is the best way to integrate new data into the digital twin?” he asks. That remains an open question, and working-group participants discussed, for example, how to integrate physics into a digital twin, which would enable it to learn and generalize more efficiently from limited data.
Some presentations at the working group explored ways to bring down the complexity of a digital twin. “When you have a model with a hundred thousand dimensions,” says Zhang, “how do you reduce it to, say, a hundred dimensions without losing too much accuracy?”
Other talks and conversations at the working group focused on specific applications — like using graph theory to predict how students might move through educational systems — or on ways to design digital-twin approaches that can be adaptable to new conditions. “The hope is that these approaches can be applied to many different systems with very little effort,” says Zhang.
Conversations at the working group have already inspired new potential collaborations that could help resolve some of the emerging issues around digital twins, says Zhang. “A lot of things I learned will keep me thinking for a long time.”
Read more about the working group "Dynamical Systems and Graph Theory Approaches for Digital Twins"