1: \begin{abstract}

2: Beyond convergence studies and comparison of different codes, there

3: are essentially no controls on the accuracy in the non-linear regime

4: of cosmological N body simulations, even in the dissipationless

5: limit. We propose and explore here a simple test which has not been

6: previously employed: when cosmological codes are used to simulate an

7: {\it isolated} overdensity, they should reproduce, in physical

8: coordinates, those obtained in open boundary conditions without

9: expansion. {In particular, the desired collisionless nature of the

10:   simulations can be probed by testing for stability in physical

11:   coordinates of virialized equilibria}.  We investigate and

12: illustrate the test using a suite of simulations in an Einstein de

13: Sitter cosmology from initial conditions which rapidly settle to

14: virial equilibrium.  We find that the criterion of stable clustering

15: allows one to determine, for given particle number $N$ in the ``halo"

16: and force smoothing $\varepsilon$, a maximum red-shift range over

17: which the collisionless limit may be represented with desired

18: accuracy. {{We also compare our results to the so-called Layzer Irvine

19:     test, showing that it provides a weaker, but very useful, tool to

20:     constrain the choice of numerical parameters.}  Finally we outline

21:   in some detail how these methods could be employed to test the

22:   choice of the numerical parameters used in a cosmological

23:   simulation.}

24: \end{abstract}

25: