1: \begin{abstract}
2: This paper considers the non-Hermitian Zakharov-Shabat (ZS) scattering problem
3: which forms the basis for defining the SU$(2)$-nonlinear Fourier transformation
4: (NFT). The theoretical underpinnings of this generalization of the conventional Fourier
5: transformation is quite well established in the Ablowitz-Kaup-Newell-Segur (AKNS)
6: formalism; however, efficient numerical algorithms that could be employed in
7: practical applications are still unavailable.
8:
9: In this paper, we present a unified
10: framework for the forward and inverse NFT using exponential one-step methods
11: which are amenable to FFT-based fast polynomial arithmetic. Within this discrete
12: framework, we propose a fast Darboux transformation (FDT) algorithm having
13: an operational complexity of $\bigO{KN+N\log^2N}$ such that the error in
14: the computed $N$-samples of the $K$-soliton vanishes as $\bigO{N^{-p}}$ where $p$
15: is the order of convergence of the underlying one-step method. For fixed $N$,
16: this algorithm outperforms the the classical DT (CDT) algorithm which has a complexity of
17: $\bigO{K^2N}$. We further present extension of these algorithms to the
18: general version of DT which allows one to add solitons to arbitrary profiles
19: that are admissible as scattering potentials in the ZS-problem. The general CDT/FDT
20: algorithms have the same operational complexity
21: as that of the $K$-soliton case and the order of convergence matches that of
22: the underlying one-step method. A comparative study of these algorithms is
23: presented through exhaustive numerical tests.
24: \end{abstract}
25: