Novel Adaptive softening for collisionless N-body simulations: Eliminating spurious halos

We describe a NOVel form of Adaptive softening (NovA) for collisionless $N$-body simulations, implemented in the Ramses adaptive mesh refinement code. We introduce a refinement criterion that the particle distribution within each cell be sufficiently isotropic, as measured by its moment of inertia tensor. In this way, collapse is only refined if it occurs along all three axes, ensuring that the softening $____epsilon$ is always of order twice the largest inter-particle spacing in a cell. This more conservative force softening criterion is designed to minimise spurious two-body effects, while maintaining high force resolution in collapsed regions of the flow. We test NovA using an antisymmetric perturbed plane wave collapse (`Valinia' test) before applying it to warm dark matter (WDM) simulations. For the Valinia test, we show that -- unlike the standard $N$-body method -- NovA produces no numerical fragmentation while still being able to correctly capture fine caustics and shells around the collapsing regions. For the WDM simulations, we find that NovA converges significantly more rapidly than standard $N$-body, producing little or no spurious halos on small scales. We show, however, that determining whether or not halos exist below the free streaming mass $M_{____rm fs}$ is complicated by the fact that our halo finder (AHF) likely incorrectly labels some caustics and criss-crossing filaments as halos, while one or two particularly massive filaments appear to fragment in any version of NovA where refinement is allowed. Such massive filaments may be physically unstable to collapse, as is the case for infinite, static, self-gravitating cylinders. We will use NovA in forthcoming papers to study the issue of halo formation below $M_{____rm fs}$; filament stability; and to obtain new constraints on the temperature of dark matter