1: \begin{abstract}

2: Single-mode optical fibers (SMFs) have become the backbone of modern

3: communication systems. However, their throughput is expected to reach

4: its theoretical limit in the nearest future. Utilization of multimode

5: fibers (MMFs) is considered as one of the most promising solutions

6: rectifying this capacity crunch. Nevertheless, differential equations

7: describing light propagation in MMFs are a way more sophisticated

8: than those for SMFs, which makes numerical modelling of MMF-based

9: systems computationally demanding and impractical for the most part

10: of realistic scenarios. Physics-informed neural networks (PINNs) are

11: known to outperform conventional numerical approaches in various domains

12: and have been successfully applied to the nonlinear Schr\"odinger

13: equation (NLSE) describing light propagation in SMFs. A comprehensive

14: study on application of PINN to the multimode NLSE (MMNLSE) is still

15: lacking though. To the best of our knowledge, this paper is the first

16: to deploy the paradigm of PINN for MMNLSE and to demonstrate that

17: a straightforward implementation of PINNs by analogy with NLSE does

18: not work out. We pinpoint all issues hindering PINN convergence and

19: introduce a novel scaling transformation for the zero-order dispersion

20: coefficient that makes PINN capture all relevant physical effects.

21: Our simulations reveal good agreement with the split-step Fourier

22: (SSF) method and extend numerically attainable propagation lengths

23: up to several hundred meters. All major limitations are also highlighted.

24: \end{abstract}

25: