About IAPA
Our Mission
To develop a realistic and quantitative understanding of how planets form, how planetary systems evolve, and how habitable planets can be identified and detected.
Planet formation is a multi-stage process. Dust particles in a protoplanetary disk grow to planetesimals, planetesimals grow to planetary embryos, and embryos collide and form planets. After the planets are formed, their orbits continue to evolve through their gravitational interactions with one another, with the remaining small bodies and, in binary systems, with the secondary star. Whether a planet can become habitable depends on all of these stages, from where it formed and how much water it received, to whether its orbit has stayed stable for billions of years.
Theoretical modeling, numerical simulations and observational campaigns form the backbone of research in planetary astrophysics. However, the long run-times of numerical simulations can heavily limit their application. A typical simulation of terrestrial planet formation takes several months to a year to complete, and modeling the transport of water and other compounds to the accretion zone of terrestrial planets requires high-resolution SPH simulations of successive collisions among planetary embryos, a type of computation that can take years. Because a large number of such simulations is needed to obtain statistically meaningful results, progress in many areas of the field has been slow.
Machine learning offers a way forward. Trained on the results of accurate simulations, machine learning models can be used as surrogates during numerical integrations to accelerate the computations, and as probabilistic estimators that predict the outcome of a simulation. They can also be used to explore large parameter spaces, and to process the large volume of data that is produced daily by ground- and space-based telescopes.
At IAPA, we study the formation, evolution, habitability and detection of planets as one connected problem, and we use machine learning as a tool in service of that science. Used in this way, machine learning allows us to address questions such as:
- Which properties of protoplanetary disks most strongly control the resulting planetary systems?
- Given a formation pathway, which regions of the parameter space produce the observed diversity of planetary masses, compositions and orbital architectures?
- For a given system, which regions of the parameter space correspond to long-term stability?
- Which physical parameters control the probabilities of collisions, ejections, capture or survival?
To this end, the Institute is committed to
- Research on planet formation, the dynamical evolution of planetary systems, habitability, and the detection of habitable and extrasolar planets.
- Developing machine learning models that are trained and tested on accurate simulations, and that can serve as reliable replacements for long and computationally expensive integrations.
- Collaboration among scientists from different institutions and countries, including theorists, observers and specialists in machine learning.
- Training students and early-career researchers in both planetary astrophysics and modern computational methods.
- Making our methods and results available to the community.