Fast Fiber Assign: Emulating Fiber Assignment Effects for Realistic DESI Catalogs
Sikandar Hanif · Deep Blue (University of Michigan) · 2025
The data-driven science of modern cosmology relies on galaxy surveys to source the millions of data points that turn targets in the sky into numbers that hold valuable information. The Dark Energy Spectroscopic Instrument (DESI), with 5000 optical fibers, is one such survey with the overarching goal of a better understanding of dark energy, the mysterious component that drives the accelerated expansion of the universe. An understanding of the various systematics that form the DESI pipeline is crucial for a better understanding of the data itself and for the ability to generate mock data catalogs with survey realism. One such step is fiber-assignment, the process of selecting which targets on the sky are observed by the instrument, especially when fiber collisions are present. When not properly accounted for, this suppresses the observed clustering amplitude. One way to simulate fiber-assignment, which is a complex and computationally expensive process, is to use an emulator, an algorithm which learns from reference datasets with the purpose of being able to emulate the effects of a process approximately but quickly. We specifically discuss the Fast Fiber Assign (FFA) algorithm which was employed by DESI to obtain clustering covariance matrices for the Data Release 1 (DR1) analysis. It was observed that the emulated covariances were not always a perfect match with analytical methods and that this could potentially be ascribed to the FFA process. Thus we especially focus on studying how tuning the emulator affects the outputs, particularly when looking at the effects on covariance matrices. We find that small adjustments to the learned kernel can be beneficial. This kernel, which is a function of the counts of close neighbors a galaxy on the sky has and how many times it could possibly be observed by the instrument, can be smoothed to reduce noise. Moreover, we find that the cosmological fitting constraints tend to be sensitive to the selection of galaxies on small scales within a `collision window'. This corresponds to the minimum separation two DESI fibers can be placed at. Using these results, we recommend improved sets of emulator parameters for the next such DESI analysis.