Regularized Maximum Likelihood Techniques for ALMA Observations 后印本

摘要: Regularized Maximum Likelihood (RML) techniques are a class of image synthesis methods that have the potential to achieve improved angular resolution and image fidelity compared to traditional image synthesis methods like CLEAN when applied to sub-mm interferometric observations. We used the GPU-accelerated open source Python package MPoL to explore the influence of various RML prior distributions (maximum entropy, sparsity, total variation, and total squared variation) on images reconstructed from ALMA continuum observations of the protoplanetary disk hosted by HD 143006. We developed a K-fold process for the image validation procedure cross-validation (CV) and explored both uniform and "dartboard" styles of visibility sampling within the validation process. Using CV to find optimal hyperparameter values for the test case of total squared variation regularization, we discovered that a wide range of hyperparameter values (spanning roughly an order of magnitude) correspond to models with strong predictive power for visibilities across unsampled or sparsely sampled spatial frequencies. We also provide a comparison of RML and CLEAN images for the protoplanetary disk around HD 143006, finding that RML imaging improves the spatial resolution of the image by about a factor of 3. Lastly, we distill general recommendations for building an RML workflow for image synthesis of ALMA protoplanetary disk observations, including recommendations for incorporating CV most effectively. Using RML methods to improve the resolution of protoplanetary disk observations will enable new science requiring high resolution images, including the detection of protoplanets embedded within disks.