Speaker
Description
State-of-the-art numerical simulations of the early stages of planet formation are often limited by computational cost due to the complexity of the included physics. Multi-physics 3D simulations of the dynamics of gas coupled with dust grains of multiple sizes including collisional evolution has long been unfeasible using conventional methods. We propose the use of machine learning, specifically neural differential equations, to train surrogate models for one component of the physics, namely the collisional growth and fragmentation of dust. With this method, we aim for a computational speed-up that would allow embedding into polydisperse hydrodynamical simulations of protoplanetary disks, and would provide greater insights into the early stages of planet formation.
| Talk category | Splinter 6: Data science |
|---|---|
| Second preference | NOVA Network 2 |
