Systematic errors of L\"uscher's fermion method and its extensions

We study the systematic errors of L\"uscher's formulation of dynamical Wilson quarks and some of its variants, in the weak and strong coupling limits, and on a sample of small configurations at finite $\beta$. We confirm the existence of an optimal window in the cutoff parameter $\varepsilon$, and the exponential decrease of the error with the number of boson families. A non-hermitian variant improves the approximation further and allows for an odd number of flavors. A simple and economical Metropolis test is proposed, which makes the algorithm exact.