Brian Barch, Daniel Lidar (Jun 05 2025).

Abstract: We analyze the computational power of non-Hermitian quantum dynamics, i.e., conditional time evolutions that arise when a quantum system is monitored and one postselects on a particular measurement record. We establish an approximate equivalence between post-selection and arbitrary non-Hermitian Hamiltonians. Namely, first we establish hardness in the following sense: Let $U=e^{-iHt}$ be an NH gate on $n$ qubits whose smallest and largest singular values differ by at least $2^{-\text{poly}(n)}$. Together with any universal set of unitary gates, the ability to apply such a gate lets one efficiently emulate postselection. The resulting model decides every language in PostBQP; hence, under standard complexity conjectures, fully scalable NH quantum computers are unlikely to be engineered. Second, we establish upper bounds which show that conversely, any non-Hermitian evolution can be written as a unitary on a system-meter pair followed by postselecting the meter. This ``purification'' is compact -- it introduces only $O(\delta^{2})$ Trotter error per time step $\delta$ -- so any NH model whose purification lies in a strongly simulable unitary family (e.g., Clifford, matchgate, or low-bond-dimension tensor-network circuits) remains efficiently simulable. Thus, non-Hermitian physics neither guarantees a quantum advantage nor precludes efficient classical simulation: its complexity is controlled by the singular-value radius of the evolution operator and by the structure of its unitary purification.

Arxiv: https://arxiv.org/abs/2506.03435