Numerical cost comparison with partial averaging

Determine whether the structural advantage of using a single Koopman finite-window generator instead of the branch structure and tuned switching criterion of partial averaging also yields a numerical advantage at matched accuracy, by comparing the computational costs of the two schemes.

Background

The paper contrasts its Koopman finite-window construction with partial averaging. The Koopman approach uses one operator-based generator across resonance crossings, whereas partial averaging uses a second transformation near a crossing and a criterion for switching between schemes.

The authors establish a structural distinction but do not determine whether that distinction reduces computational cost when both methods are required to achieve the same accuracy. A matched-accuracy numerical comparison is therefore left unresolved.