Amortized Clustering Assistant Classification of Anomalous Hybrid Floquet Modes in a Periodically Driven non-Hermitian Lattice

Abstract

The interplay between Floquet periodically driving and non-Hermiticity could bring about intriguing novel phenomena with anomalous Floquet topological phases of a finite-size, tight-binding lattice model. How to efficiently investigate on quasi-energy and eigenfield of a non-Hermitian Floquet system with complicated driving protocol remains a challenging task. In this work, we define a somewhat complex driving protocol for a bipartite lattice system and discover two nontrivial topological phases that support Floquet {\pi} mode. Thereafter, we introduce unsupervised learning method in order to explore distribution features of system eigenfunctions under different magnitude of system energy gain/loss. We utilize the idea of amortized clustering and construct an algorithm selector that could dynamically upgrade with increasing gain/loss as input parameter. Proper employment of the selector enables us to reveal the regulation of dynamic localization from abundant possible wave function distribution in two-dimension lattice in another efficient way. In addition, our work provides a feasible methodology via machine learning method to assist in classification of Floquet modes.