New metric to quantify the similarity between planetary systems: application to dimensionality reduction using T-SNE^★

Physikalisches Institut & NCCR PlanetS, Universität Bern, 3012 Bern, Switzerland
e-mail: alibert@space.unibe.ch

Received: 7 November 2018
Accepted: 24 January 2019

Abstract

Context. Planet formation models now often consider the formation of planetary systems with more than one planet per system. This raises the question of how to represent planetary systems in a convenient way (e.g. for visualisation purpose) and how to define the similarity between two planetary systems, for example to compare models and observations.

Aims. We define a new metric to infer the similarity between two planetary systems, based on the properties of planets that belong to these systems. We then compare the similarity of planetary systems with the similarity of protoplanetary discs in which they form.

Methods. We first define a new metric based on mixture of Gaussians, and then use this metric to apply a dimensionality reduction technique in order to represent planetary systems (which should be represented in a high-dimensional space) in a two-dimensional space. This allows us study the structure of a population of planetary systems and its relation with the characteristics of protoplanetary discs in which planetary systems form.

Results. We show that the new metric can help to find the underlying structure of populations of planetary systems. In addition, the similarity between planetary systems, as defined in this paper, is correlated with the similarity between the protoplanetary discs in which these systems form. We finally compare the distribution of inter-system distances for a set of observed exoplanets with the distributions obtained from two models: a population synthesis model and a model where planetary systems are constructed by randomly picking synthetic planets. The observed distribution is shown to be closer to the one derived from the population synthesis model than from the random systems.

Conclusions. The new metric can be used in a variety of unsupervised machine learning techniques, such as dimensionality reduction and clustering, to understand the results of simulations and compare them with the properties of observed planetary systems.