Partial Observation of Linear Systems with the Mori-Zwanzig Formalism
Abstract
The Mori-Zwanzig formalism provides a systematic framework for deriving reduced-order model of dynamical systems when only part of the state is observed, but its practical use is often limited by the complexity of the resulting computations. This paper develops an explicit formulation of the Mori-Zwanzig equation for linear time-invariant systems under partially observed observables. By expressing the dynamics in terms of observables, the Koopman generator, and projections onto resolved and unresolved components, we derive closed-form representations of the Markovian, noise, and memory contributions that arise in the Mori-Zwanzig identity. For the linear setting, the resulting formulas recover the reduced dynamics obtained from the variation-of-constants formula while retaining the operator-based structure of the Mori-Zwanzig approach. This makes the derivation a transparent reference case for reduced-order modelling with memory and clarifies how unresolved variables influence the observed dynamics through history-dependent terms. The analysis also identifies the ingredients needed for extensions to nonlinear systems and more general projections, including spectral filtering and data-driven approximations of memory effects. Analytical and numerical examples involving the harmonic oscillator and wave equations illustrate the construction and demonstrate how the formalism can be used to obtain interpretable reduced-order models for partially observed systems.
Versions
Related
- Physics-informed traditional and generative Mori-Zwanzig-reduced models for multiscale systems
- Koopman-von Neumann formalism meets contact Hamiltonian systems
- Thermodynamic formalism for non-compact systems with expansivity and specification
- Identification and Recovery of Linear Dynamics from Partial Trajectories
- Convergence Analysis of Variable-Separation-Based Model Reduction for Linear Parametric Partial Differential Equations
BibTeX
@misc{wang2026partial,
title = {{Partial Observation of Linear Systems with the Mori-Zwanzig Formalism}},
author = {Fan Wang and Peter Benner and Jan Heiland},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2606.23341},
eprint = {2606.23341},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.23341},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Partial Observation of Linear Systems with the Mori-Zwanzig Formalism
AU - Fan Wang
AU - Peter Benner
AU - Jan Heiland
PY - 2026
DA - 2026/10/31
PB - arXiv
DO - 10.48550/arxiv.2606.23341
UR - https://arxiv.org/abs/2606.23341
AB - The Mori-Zwanzig formalism provides a systematic framework for deriving reduced-order model of dynamical systems when only part of the state is observed, but its practical use is often limited by the complexity of the resulting computations. This paper develops an explicit formulation of the Mori-Zwanzig equation for linear time-invariant systems under partially observed observables. By expressing the dynamics in terms of observables, the Koopman generator, and projections onto resolved and unresolved components, we derive closed-form representations of the Markovian, noise, and memory contributions that arise in the Mori-Zwanzig identity. For the linear setting, the resulting formulas recover the reduced dynamics obtained from the variation-of-constants formula while retaining the operator-based structure of the Mori-Zwanzig approach. This makes the derivation a transparent reference case for reduced-order modelling with memory and clarifies how unresolved variables influence the observed dynamics through history-dependent terms. The analysis also identifies the ingredients needed for extensions to nonlinear systems and more general projections, including spectral filtering and data-driven approximations of memory effects. Analytical and numerical examples involving the harmonic oscillator and wave equations illustrate the construction and demonstrate how the formalism can be used to obtain interpretable reduced-order models for partially observed systems.
ER -