Improvements to ATLAS primary vertex reconstruction for LHC Run 3

Increasing pileup multiplicity in ATLAS, reaching 70 inelastic proton-proton collisions per beam crossing for Run-2 of the LHC, and even higher in future runs, poses challenges for primary vertex reconstruction. To address these challenges, two new tools were adapted. The first, a Gaussian Track Density Seed Finder, uses a simple yet powerful analytic model of track density along the beam axis to locate candidate vertex positions. The second, an Adaptive Multi Vertex Finder, employs a global approach to vertex finding and fitting, and extends the adaptive vertex fitting technique used in Run-2 by allowing nearby vertices to compete for compatible tracks. This talk summarises the optimization and expected performance of the primary vertex reconstruction software under conditions expected for Run-3 of the LHC.