Stability mapping of bipartite tight-binding graphs with losses and gain: ${\cal PT}-$symmetry and beyond
We consider bipartite tight-binding graphs composed by $N$ nodes split into two sets of equal size: one set containing nodes with on-site loss, the other set having nodes with on-site gain. The nodes are connected randomly with probability $p$. We give a rationale for the relevance of such "throttle/brake" coupled systems (physically open systems) to grasp the stability issues of complex networks in areas such as biochemistry, neurons or economy, for which their modelling in terms of non-hermitian Hamiltonians is still in infancy. Specifically, we measure the connectivity between the two sets with the parameter $α$, which is the ratio of current adjacent pairs over the total number of possible adjacent pairs between the sets. For general undirected-graph setups, the non-hermitian Hamiltonian $H(γ,α,N)$ of this model presents pseudo-Hermiticity, where $γ$ is the loss/gain strength. However, we show that for a given graph setup $H(γ,α,N)$ becomes ${\cal PT}-$symmetric. In both scenarios (pseudo-Hermiticity and ${\cal PT}-$symmetric), depending on the parameter combination, the spectra of $H(γ,α,N)$ can be real even when it is non-hermitian. Thus, we numerically characterize the average fractions of real and imaginary eigenvalues of $H(γ,α,N)$ as a function of the parameter set $\{γ,α,N\}$. We demonstrate, for both setups, that there is a well defined sector of the $γα-$plane (which grows with $N$) where the spectrum of $H(γ,α,N)$ is predominantly real.