A Bayesian neural network predicts the dissolution of compact planetary systems

@article{Cranmer2021ABN,
  title={A Bayesian neural network predicts the dissolution of compact planetary systems},
  author={M. Cranmer and Daniel Tamayo and Hanno Rein and Peter W. Battaglia and Sam Hadden and Philip J. Armitage and Shirley Ho and David N. Spergel},
  journal={Proceedings of the National Academy of Sciences of the United States of America},
  year={2021},
  volume={118}
}
  • M. Cranmer, D. Tamayo, D. Spergel
  • Published 11 January 2021
  • Computer Science, Physics, Geology
  • Proceedings of the National Academy of Sciences of the United States of America
Significance Despite over 300 y of effort, no solutions exist for predicting when a general planetary configuration will become unstable. We introduce a deep learning architecture to push forward this problem for compact systems. While current machine learning algorithms in this area rely on scientist-derived instability metrics, our new technique learns its own metrics from scratch, enabled by a internal structure inspired from dynamics theory. Our model can quickly and accurately predict… 

Figures and Tables from this paper

Leveraging Stochastic Predictions of Bayesian Neural Networks for Fluid Simulations
TLDR
It is shown that BNN-based corrector networks can stabilize coarse-grained simulations and successfully create multimodal trajectories and two use-cases of a combination of BNNs with fluid simulations: uncertainty estimation and multi-modal synthesis.
What Are Bayesian Neural Network Posteriors Really Like?
TLDR
It is shown that BNNs can achieve significant performance gains over standard training and deep ensembles, and a single long HMC chain can provide a comparable representation of the posterior to multiple shorter chains, and posterior tempering is not needed for near-optimal performance.
A Criterion for the Onset of Chaos in Compact, Eccentric Multiplanet Systems
We derive a semianalytic criterion for the presence of chaos in compact, eccentric multiplanet systems. Beyond a minimum semimajor axis separation, below which the dynamics are chaotic at all…
On Uncertainty, Tempering, and Data Augmentation in Bayesian Classification
TLDR
It is shown that a cold posterior, tempered by a power greater than one, often more honestly reflects the authors' beliefs about aleatoric uncertainty than no tempering, providing an explicit link between data augmentation and cold posteriors.
Machine learning as a tool in theoretical science
  • M. Douglas
  • Computer Science
    Nature Reviews Physics
  • 2022

References

SHOWING 1-10 OF 98 REFERENCES
A Machine Learns to Predict the Stability of Tightly Packed Planetary Systems
TLDR
It is found that training an XGBoost machine learning algorithm on physically motivated features yields an accurate classifier of stability in packed systems, and motivates investing computational resources to train algorithms capable of predicting stability over longer timescales and over broader regions of phase space.
Predicting the long-term stability of compact multiplanet systems
TLDR
The Stability of Planetary Orbital Configurations Klassifier (SPOCK) predicts stability using physically motivated summary statistics measured in integrations of the first 104 orbits, thus achieving speed-ups of up to 105 over full simulations, which computationally opens up the stability-constrained characterization of multiplanet systems.
Fundamental limits from chaos on instability time predictions in compact planetary systems
Instabilities in compact planetary systems are generically driven by chaotic dynamics. This implies that an instability time measured through direct N-body integration is not exact, but rather…
Dynamical instability and its implications for planetary system architecture
We examine the effects that dynamical instability has on shaping the orbital properties of exoplanetary systems. Using N-body simulations of non-EMS (Equal Mutual Separation), multi-planet systems we…
Nekhoroshev Estimates for the Survival Time of Tightly Packed Planetary Systems
$N$-body simulations of non-resonant tightly-packed planetary systems have found that their survival time (i.e. time to first close encounter) grows exponentially with their interplanetary spacing…
Bayesian Deep Learning and a Probabilistic Perspective of Generalization
TLDR
It is shown that deep ensembles provide an effective mechanism for approximate Bayesian marginalization, and a related approach is proposed that further improves the predictive distribution by marginalizing within basins of attraction, without significant overhead.
Deep learning of multi-element abundances from high-resolution spectroscopic data
  • Henry W Leung, J. Bovy
  • Computer Science, Physics
    Monthly Notices of the Royal Astronomical Society
  • 2018
TLDR
This work designs a neural network for high-resolution spectroscopic analysis using APOGEE data that mimics the methodology of standard spectroscopy analyses: stellar parameters are determined using the full wavelength range, but individual element abundances use censored portions of the spectrum.
THE STATISTICAL MECHANICS OF PLANET ORBITS
The final ?giant-impact? phase of terrestrial planet formation is believed to begin with a large number of planetary ?embryos? on nearly circular, coplanar orbits. Mutual gravitational interactions…
The Stability of Multi-Planet Systems
A system of two small planets orbiting the Sun on low-eccentricity, low-inclination orbits is stable with respect to close encounters if the initial semi-major axis difference, Ξ”, measured in mutual…
Hill stability in the AMD framework
In a two-planet system, a topological boundary that is created by Sundman (1912, Acta Math., 36, 105) inequality can forbid close encounters between the two planets for an infinite time. A system is…
...
1
2
3
4
5
...