We study the small scale structure of axion dark matter in the post-inflationary scenario, which predicts the formation of low-mass, high density clumps of gravitationally bound axions called axion miniclusters. To this end we follow numerically the cosmological evolution of the axion field and the network of strings and domain walls until the density contrast is frozen. Our simulations, comprising up to 8192 3 points, are the largest studies of the axion field evolution in the non-linear regime presented so far. Axitons, pseudobreathers of the Klein-Gordon equation, are observed to form in our simulation at late times. Studying their properties analytically and numerically, we observe that in particular the earliest axitons contribute to density perturbations at the typical length scale of miniclusters. We analyse the small scale structure of the density field, giving the correlation length, power spectrum and the distribution of high density regions that will collapse into axion miniclusters. The final density field of our simulations can be used to calculate the minicluster mass fraction in simulations including gravity. In particular, we find that typical minicluster progenitors are smaller than previously thought and only of moderate, O(1) overdensity. We expect these miniclusters to have a rich sub-structure, emerging from small-scale fluctuations produced in the collapse of the string-wall network and from axitons. arXiv:1809.09241v2 [astro-ph.CO] 9 Apr 2019 3 In the low T regime the potential away from the minimum is not well-described by (2.2). However, it seems to be a good approximation above T ∼ 2Tc or, at least, the quartic coupling computed in [68] does. The fact that at these temperatures the axion field is mostly very close to θ ∼ 0 renders the inaccuracy of small importance to us in this paper, but, as we will see, it is not entirely irrelevant.4 Definitions in the literature differ slightly, for instance Sikivie and Kawasaki use 3H1 = mA as the defining scale. Fortunately for comparison's sake, this has only mild effect on the value of H1 as the temperature dependence of mA is quite strong.