1: \begin{abstract}

2: Optimal balance is a non-asymptotic numerical method for computing a

3: point on an elliptic slow manifold for two-scale dynamical systems

4: with strong gyroscopic forces.  It works by solving a modified

5: differential equation as a boundary value problem in time, where the

6: nonlinear terms are adiabatically ramped up from zero to the fully

7: nonlinear dynamics.  A dedicated boundary value solver, however, is

8: often not directly available.  The most natural alternative is a

9: nudging solver, where the problem is repeatedly solved forward and

10: backward in time and the respective boundary conditions are restored

11: whenever one of the temporal end points is visited.  In this paper, we

12: show quasi-convergence of this scheme in the sense that the

13: termination residual of the nudging iteration is as small as the

14: asymptotic error of the method itself, i.e., under appropriate

15: assumptions exponentially small.  This confirms that optimal balance

16: in its nudging formulation is an effective algorithm.  Further, it

17: shows that the boundary value problem formulation of optimal balance

18: is well posed up at most a residual error as small as the asymptotic

19: error of the method itself.  The key step in our proof is a careful

20: two-component Gronwall inequality.

21: \end{abstract}

22: