1: \begin{abstract}

2: We explore two classes of exponential integrators in this letter to design

3: nonlinear Fourier transform (NFT) algorithms with a desired

4: accuracy-complexity trade-off and a convergence order of $4$ on an 

5: equispaced grid. The integrating factor based method in the class of

6: Runge-Kutta methods yield algorithms with complexity $O(N\log^2N)$ (where $N$ is

7: the number of samples of the signal) which have superior accuracy-complexity

8: trade-off than any of the fast methods known currently.

9: The integrators based on Magnus series expansion, namely, standard and 

10: commutator-free Magnus methods yield algorithms of complexity $O(N^2)$ that 

11: have superior error behavior even for moderately small step-sizes and higher 

12: signal strengths.

13: \end{abstract}

14: