2018MNRAS.480..371L

The degree of fractal substructure in molecular clouds can be quantified by comparing them with fractional Brownian motion (FBM) surfaces or volumes. These fields are self-similar over all length-scales and characterized by a drift exponent H, which describes the structural roughness. Given that the structure of molecular clouds and the initial structure of star clusters are almost certainly linked, it would be advantageous to also apply this analysis to clusters. Currently, the structure of star clusters is often quantified by applying Q analysis. Q values from observed targets are interpreted by comparing them with those from artificial clusters. These are typically generated using a box-fractal (BF) or radial density profile (RDP) model. We present a single cluster model, based on FBM, as an alternative to these models. Here, the structure is parametrized by H and the standard deviation of the log-surface/volume density σ. The FBM model is able to reproduce both centrally concentrated and substructured clusters, and is able to provide a much better match to observations than the BF model. We show that Q analysis is unable to estimate FBM parameters. Therefore, we develop and train a machine learning algorithm that can estimate values of H and σ, with uncertainties. This provides us with a powerful method for quantifying the structure of star clusters in terms that relate to the structure of molecular clouds. We use the algorithm to estimate the H and σ for several young star clusters, some of which have no measurable BF or RDP analogue.