Title: Defining subsystems in Hilbert spaces with non-Euclidean metric

Abstract: This work outlines a consistent method of identifying subsystems in finite-dimensional Hilbert spaces, independent of the underlying inner-product structure. It has been well established that Hilbert spaces with modified inner-product, defined through the so-called metric operator, turn out to be the most natural ways to represent certain phenomena such as those involving balanced gain and loss resulting in pseudo-Hermitian Hamiltonians. For composite systems undergoing pseudo-Hermitian evolution, defining the subsystems is generally considered feasible only when the metric operator is chosen to have a tensor product form so that a partial trace operation can be well defined. In this work, we use arguments from algebraic quantum mechanics to show that the subsystems can be well-defined in every metric space -- irrespective of whether or not the metric is of tensor product form. This is done by identifying subsystems with a decomposition of the underlying $C^*$-algebra into commuting sub-algebras. We show that different subsystem decompositions correspond to choosing different equivalence classes of the GNS representation. Furthermore, given a form of pseudo-Hermitian Hamiltonian, the choice of the Hamiltonian compatible metric characterizes the subsystem decomposition and as a consequence, the entanglement structure in the system. We clarify how each of the subsystems, defined this way, can be tomographically constructed and that these subsystems satisfy the no-signaling principle. With these results, we put all the choices of the metric operator on an equal footing.