Improving simulations of mantle convection within rocky planets
Improving simulations of mantle convection within rocky planets lead image
Mantle convection is one of the key dynamic processes inside rocky planets, driving volcanism, plate tectonics, and magnetic field generation. Understanding mantle convection, through simulation and analysis, can provide insights into planetary formation and evolution. However, the large range of parameters and initial conditions can make running these simulations costly and time-consuming.
Agarwal et al. developed a machine learning-based approach to predict mantle convection, replacing the most computationally intensive component of traditional simulations: solving the Stokes equations.
“Our machine learning model speeds up these simulations by a factor of up to 89,” said author Siddhant Agarwal. “This means that a typical simulation that would take five days to finish can now be performed in a little over an hour.”
The researchers trained their model on 94 mantle simulations, instructing it to predict its flow velocity as a function of temperature. This data is then fed into a finite-volume solver to determine the evolution of the temperature field. The result is an approach that is much faster while maintaining accuracy, capable of modeling the thermal evolution of rocky planets’ mantles even after only being trained on steady-state snapshots.
The authors did find their model struggles when given initial conditions far outside its training data, so they plan to make the training process more robust and versatile.
“We are looking to extend our approach to different grids in two and possibly even three dimensions, as well as introduce some other additions to the input space of the model in a way that would make the trained model readily useful to the community for a wide range of studies,” said Agarwal.
Source: “Physics-based machine learning for mantle convection simulations,” by Siddhant Agarwal, Ali Can Bekar, Christian Hüttig, David S. Greenberg, and Nicola Tosi, Physics of Fluids (2025). The article can be accessed at https://doi.org/10.1063/5.0281832 .
