1: \begin{abstract}

2: In this paper, two multiscale time integrators (MTIs), motivated from two types of multiscale

3: decomposition by either frequency or frequency and amplitude, are proposed and analyzed for

4: solving highly oscillatory second order differential equations

5: with a dimensionless parameter $0<\eps\le1$. In fact, the solution to this equation

6: propagates waves with wavelength at $O(\eps^2)$ when  $0<\eps\ll 1$, which brings significantly

7: numerical burdens in practical computation. We

8: rigorously establish  two independent error bounds for the two MTIs

9: at $O(\tau^2/\eps^2)$ and $O(\eps^2)$ for $\eps\in(0,1]$ with $\tau>0$ as step size, which

10: imply that the two MTIs converge uniformly with linear convergence rate

11: at $O(\tau)$ for  $\eps\in(0,1]$ and optimally with quadratic convergence rate at $O(\tau^2)$

12: in the regimes when either $\eps=O(1)$ or $0<\eps\le \tau$.

13: Thus the meshing strategy requirement (or $\eps$-scalability) of the two MTIs

14: is $\tau=O(1)$ for $0<\eps\ll 1$,

15: which is significantly improved from $\tau=O(\eps^3)$ and $\tau=O(\eps^2)$

16: requested by finite difference methods and exponential wave integrators to

17: the equation, respectively.  Extensive numerical tests and comparisons with

18: those classical numerical integrators are reported, which gear towards better

19: understanding  on the convergence and resolution properties of  the two MTIs.

20: In addition, numerical results support the two error bounds very well.

21: \end{abstract}

22: