Simulating full QCD with the fixed point action

Because of its complex structure the parametrized fixed point action can not be simulated with the available local updating algorithms. We constructed, coded, and tested an updating procedure with $2+1$ light flavors, where the targeted $s$ quark mass is at its physical value while the $u$ and $d$ quarks should produce pions lighter than 300 MeV. In the algorithm a partially global gauge update is followed by several accept/reject steps, where parts of the determinant are switched on gradually in the order of their costs. The trial configuration that is offered in the last, most expensive, stochastic accept/reject step differs from the original configuration by a Metropolis $+$ over-relaxation gauge update over a subvolume of $\sim (1.3\mathrm{fm}{)}^4$. The acceptance rate in this accept/reject step is $\sim 0.4$. The code is optimized on different architectures and is running on lattices with $L_s\approx 1.2\mathrm{fm}$ and 1.8 fm at a resolution of $a\approx 0.15\mathrm{fm}$.