arXiv:2610.29316
Boundary exceptional point driven chiral switch of topological edge states
Bikashkali Midya, Barsa Rani Sahu
Abstract
We theoretically investigate the interplay of topology and localized non-Hermiticity in a finite Su-Schrieffer-Heeger chain characterized by non-Hermitian potentials at its boundaries. By tuning the boundary gain and loss parameters, we exploit the finite-size hybridization of two mid-gap topological edge states to analytically derive the exact coordinates of boundary-induced exceptional points (EPs), $Γ_{\rm EP}=\pm i v^{N}/w^{N-1}$, at which these states undergo coalescence. We demonstrate that dynamically encircling these boundary EPs facilitates a robust, chiral edge state-transfer protocol. Depending on the orientation of the encircling loop, this protocol allows a topological edge state localized at one boundary to be either transferred to the opposite edge with perfect fidelity or returned to its starting position through a double switching process. Importantly, since the non-Hermiticity remains strictly confined to the boundaries, this chiral transport occurs without affecting the energy spectrum of the bulk bands. These results establish a minimal framework for a large-scale topological chiral switch that is experimentally verifiable and capable of being controlled through boundary non-Hermitian driving.
AI Summary
The paper asks whether chiral exceptional-point (EP) state switching can transfer topological edge excitations across an SSH chain using only boundary controls, without modulating the bulk. It studies $N$ unit cells with fixed alternating hoppings $v<w$ and opposite endpoint potentials $\pm\Gamma(t)$, where $\Gamma=\delta+i\gamma$ controls detuning and gain/loss. Zero-energy spatial recurrence relations give the exact boundary-induced EP locations, $\Gamma_{\rm EP}=\pm i v^N/w^{N-1}$, where the two finite-size-hybridized edge modes coalesce. A projected two-edge-state Hamiltonian explains the competition between exponentially small interedge coupling and boundary gain/loss. Time-dependent Schrödinger evolution is then simulated for circular parameter loops enclosing both EPs, starting and ending at Hermitian configurations with detuning larger than the edge coupling. The simulations show direction-dependent selection of the final edge: one loop orientation transfers an initially left-localized excitation rightward, while the reverse orientation returns it leftward through two switches. The final selected edge is reported to be independent of which boundary is initially excited. A representative calculation uses 16 sites, $w/v=1.8$, loop radius $0.1v$, and duration $200/v$; the EPs lie at $\gamma/v\simeq\pm0.016$. A numerical size–duration scan reports an approximately linear minimum duration for high transfer. The analysis assumes weak boundary gain/loss relative to the bulk gap and relies on finite-size edge hybridization. Transfer is quantified through normalized sublattice populations. The results are theoretical; experimental implementations and extensions to Liouvillian dynamics and higher-dimensional systems remain prospective.
Central claim may not hold: The claimed unchanged bulk-band spectrum is false for the finite-chain Hamiltonian, although the static infinite-bulk Hamiltonian and its topological invariant can remain unchanged. The reported sublattice population is not target-edge-state fidelity, and the single-site initialization already overlaps bulk modes, contradicting the assertion that bulk bands remain unpopulated. Exact input-independent perfect edge-state selection is incompatible with the invertible finite-time propagator, whereas approximate normalized mode selection remains possible. The effective two-state projection omits the edge-vector normalization factor, preserving the EP coordinates but misnormalizing effective energies and dynamical rates. Figure 3's caption contradicts its own row labels and the main text about which loop orientation produces left-to-right transfer.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.