IGyTheM: Integrating Geodynamics with Thermodynamics using Machine learning

To achieve a comprehensive understanding of mantle convection, it is essential to develop models that incorporate the complex interactions between chemical structures and mantle dynamics. However, incorporating thermodynamic calculations in a self-consistent and computationally efficient manner remains a major challenge in computational geodynamics. Two conventional approaches are commonly used to couple geodynamic models with thermodynamic properties. The first relies on pre-computed lookup tables derived from thermodynamic calculations at fixed bulk compositions (e.g., Li et al., 2025; Rummel et al., 2020), which limits compositional flexibility and resolution. The second approach performs thermodynamic calculations on-the-fly during geodynamic simulations(e.g., Hebert et al., 2009), providing greater physical consistency but at a computational cost that is often prohibitive, including approaches based on parallelized, parameterized thermodynamic calculations (e.g., Riel et al., 2022; Wong & Keller, 2023).

Here, we present a new, flexible, general, and computationally efficient framework for coupling convection models with thermodynamics, termed Integrating Geodynamics with Thermodynamics using Machine Learning (IGyTheM, Fig. 1) (Yuan et al., 2025). Assuming ... [full article].

Contributed by: Qian Yuan, Texas A&M University, College Station, Texas, USA ; Paul D. Asimow, Michael Gurnis, Paula Antoshechkina, Junjie Dong, California Institute of Technology, Pasadena, California, USA

Check out all Research Highlights