We evaluate the efficiency of axion production from spatially random initial conditions in the axion field, so a network of axionic strings is present. For the first time, we perform numerical simulations which fully account for the large short-distance contributions to the axionic string tension, and the resulting dense network of high-tension axionic strings. We find nevertheless that the total axion production is somewhat less efficient than in the angle-averaged misalignment case. Combining our results with a recent determination of the hot QCD topological susceptibility (Borsanyi et al 2016), we find that if the axion makes up all of the dark matter, then the axion mass is m_a = 26.2 +-3.4 micro-electron volts.

Global string networks may be relevant in axion production in the early Universe, as well as other cosmological scenarios. Such networks contain a large hierarchy of scales between the string core scale and the Hubble scale, log(f/H) around 70, which influences the network dynamics by giving the strings large tensions T = pi f^2 log(f/H). We present a new numerical approach to simulate such global string networks, capturing the tension without an exponentially large lattice.