Work

Fast Propagation of Normal Dissipative Dynamics

Year: 2026
Type: preprint
Abstract: When a generator makes a system oscillate and decay quickly, a direct time-stepping calculation can spend most of its effort resolving motion that never appears in the final readout. This paper gives a deterministic algorithm for the full endpoint of an explicitly supplied rational matrix evolution. If the leading matrix is normal and dissipative at each time, the cost of its oscillation and damping scales is polynomial in their binary encoding lengths, even at exact spectral crossings. The cost of an additional polynomial perturbation depends on its numerical norm bound. The algorithm groups nearby eigenvalues and transports the resulting spectral subspaces. It evaluates the rapidly varying scalar factors in the forward direction, with a separate error budget for their interaction with the remaining block dynamics. For permutation-equivariant Lindblad dynamics at fixed local dimension, an exact rational representation gives a reusable channel description. It supports classical sampling of the entire labelled measurement string to a specified total variation error for site-dependent product preparations and measurements, including a binary-encoded number of repeated drive periods. Spatial locality alone gives a different complexity bound. A single-site probability is PSPACE-hard to estimate at an exponentially long time given in binary, even for a fixed-dimensional nearest-neighbor chain with self-adjoint, unital, purely dissipative dynamics. The reduction uses a designated invariant component. A worked two-level example certifies how dephasing changes the readout at a narrow avoided crossing.
Language: English
Open Access status: green
OpenAlex ID: W7236206864
Cites:
Cited by:
Related to: 10