Partial Impulse Framework

• Partial Impulse Framework is a set of mathematical and computational methods for systems evolving continuously with discrete impulses under limited observability.
• It integrates techniques like filtering, change-of-measure, and dynamic programming to address impulse control in stochastic and hybrid systems, enhancing optimal policy design.
• The framework extends to observability, partial resets, and physical threshold estimation, providing robust approaches for practical problems such as resource management and fluid simulations.

The partial impulse framework designates a class of mathematical and computational methodologies that address systems subject to both continuous evolution and discrete, instantaneous changes—impulses—where only partial information about the system state, control, observability, or effect is available or relevant. This paradigm appears in impulse control with partial observations, hybrid control design with partial reset or delayed execution, and physical threshold estimation where only sub-threshold or partially observable effects are accessible. Rigorous partial impulse frameworks enable optimal policy construction, state estimation, or stability analysis under these nonclassical conditions.

1. Partial Impulse Control in Stochastic Systems

Partial impulse control in stochastic systems refers to decision problems in which a controller enacts instantaneous interventions based on filtered or partially observed information. A prototypical example is impulse control of a diffusion process whose drift parameter is unknown and subject to abrupt, unobservable change-points. The framework employs filtering (posterior estimation) and change-of-measure arguments to recast the original partially observed problem into a higher-dimensional, fully-observed Markov process. Specifically, posterior likelihoods and expectations are tracked via SDEs driven by innovation processes, leading to an augmented Markov state for which standard impulse control theory (dynamic programming, quasi-variational inequalities) applies. The resulting solution characterizes optimal impulse times and sizes in terms of explicit value functions and intervention operators, and supports numerical schemes such as Longstaff–Schwartz regression-based Monte Carlo for high-dimensional problems (Abbas-Turki et al., 2014).

2. Impulse Control in Piecewise Deterministic and Partially Observed Regimes

Piecewise Deterministic Markov Processes (PDMPs) with partially observed modes or states provide a canonical domain for the partial impulse framework. The control objective is to minimize expected cumulative costs for systems whose discrete modes or jump-times are hidden and for which interventions are only allowed at randomly timed or noisy observation events. The methodology maps the problem onto a Partially Observable Markov Decision Process (POMDP) with a continuous or hybrid state space, using belief-state (filter) dynamics to propagate posterior distributions over system modes. A two-stage discretization algorithm—first of the PDMP state space, then of the belief space—permits convergence-guaranteed finite-state dynamic programming. The approach accommodates practical challenges such as boundary-triggered jumps, state aggregation, and reveals near-optimal control strategies validated on clinically relevant models (e.g., cancer relapse treatment policies) (Cleynen et al., 2021).

3. Partial Impulse Control with Observation and Execution Delays

For stochastic systems with non-smooth or regime-switching dynamics, partial impulse control entails policies enacted at random, discrete observation times and subjected to stochastic delays between decision and execution. The resulting value function must incorporate both the running cost (e.g., for depletion or undesired regimes) and the anticipated cost due to delayed execution under partial information. The policy synthesis reduces to coupled degenerate elliptic PDEs—one for the pre-decision value and another auxiliary equation for delayed (post-decision, pre-execution) costs—linked via quasi-variational inequalities. The optimal control exhibits threshold-type structure, where intervention is triggered if the observed or filtered state falls below a regime-specific level. Increased delay or decreased observation frequency raises the thresholds, reflecting greater risk. This framework is realized in contexts such as resource replenishment in environmental systems, and supports numerical implementation via semi-Lagrangian and finite-volume discretizations (Yoshioka et al., 2020).

4. Partial Impulse Observability and State Reconstruction

The partial impulse framework extends to the problem of system observability, particularly in singular or descriptor systems where impulsive (Dirac delta) behaviors may arise due to inconsistent initial conditions or algebraic constraints. The fundamental question is whether the so-called impulsive components of an unmeasured linear combination of states (denoted z(t)) can be inferred from the measured output (y(t)), regardless of initial conditions. Formalized as partial impulse observability, the property requires that all unmeasured impulsive behavior in z is observable through y. The characterization is achieved via an algebraic rank condition on an appropriately constructed block matrix (F_{n+1}, F_{n+1,L}), or equivalently through geometric criteria based on Wong sequences. These results underlie robust observer synthesis for singular systems that must handle, but not amplify, impulsive state components (Jaiswal et al., 2022).

5. Partial Resets in Hybrid and Impulsive Dynamical Systems

In hybrid control systems—such as reset control circuits—the partial impulse framework manifests in the selective resetting (“partial reset”) of a subset of the system states upon occurrence of prescribed events (e.g., zero-crossings of a monitored variable). The hybrid dynamical system alternates between linear or nonlinear flow and impulses governed by a reset map with block structure distinguishing resetting and non-resetting states. The theoretical advances are explicit: for well-posedness (no Zeno phenomenon), existence and uniqueness of solutions, and continuous dependence on initial data, specific algebraic invariance criteria must be satisfied by the subspace of reset states. Partial resets with "right" or "left" block structures guarantee well-posedness, and the strong continuous dependence on initial conditions follows under transversal crossing assumptions. Sensor-noise robustness is secured when the augmented system (including the noise generator) maintains the same well-posed crossing property. These results support both the analysis and practical engineering of robust hybrid control systems with partial impulse actions (Baños et al., 2015).

6. Partial Impulse Thresholds in Physical Systems

In physical phenomena such as sediment grain dislodgement under hydrodynamic forces, partial impulse frameworks quantify the distinction between partial and full motion. The work–energy principle yields a scalar impulse threshold J_c: only if the time-integrated external impulse exceeds J_c will a particle traverse the unstable separatrix and become fully lodged; sub-threshold impulses produce only partial, reversible excursions. The threshold depends critically on the grain’s inertia, submerged weight, and the micro-topographic barrier (Δz). Empirical and theoretical analyses demonstrate that the closed-form root law J_c = √[2 m_eff w_s Δz] matches experimental data and recovers observed scaling relations between force magnitude, pulse duration, and dislodgement. The framework’s predictive value is evidenced in experimental design, threshold reinterpretation, and generalization to more complex geometries (1901.10210).

7. Impulse-Based Decomposition in PDEs and Navier–Stokes Characteristic Mapping

Recent developments in flow simulation leverage partial impulse decompositions to achieve unified solvers for incompressible fluid flows subject to viscosity and forcing. In this context, the "impulse" is a vector gauge quantity associated with the velocity field, decomposed into a Lie-advected (pure-transport) part along particle flow maps and a complementary source generated by viscosity and body forces, accumulated as a path integral. By tracking both components on Lagrangian particles and periodically projecting to a grid for velocity recovery, the partial impulse framework maintains high geometric fidelity and compatibility with classical solvers. This methodology provides a robust foundation for simulating fluid phenomena where standard vorticity or velocity approaches are suboptimal (Li et al., 31 Jan 2026).