Measure Differential Equations

• Measure Differential Equations are differential equations in which the unknown or the data is a measure, enabling analysis of discontinuities, impulses, and singular dynamics.
• They provide a unifying framework for classical ODEs, impulsive systems, hybrid dynamics, and stochastic filters, with applications ranging from kinetic theory to control.
• Solution techniques include Kurzweil–Stieltjes integration, renormalized methods, and Lagrangian numerical schemes, ensuring rigorous existence, uniqueness, and stability analysis.

A Measure Differential Equation (MDE) is a differential equation in which either the unknown or the data is a measure, or the evolution is described in a measure-theoretic or distributional sense. This encompasses differential equations driven by general measures, nonlinear equations on spaces of probability measures, and generalized ODEs and PDEs in the sense of Kurzweil–Stieltjes or distributions. MDEs provide a unifying framework that subsumes classical ODEs, impulsive and hybrid systems, stochastic filters, nonlocal kinetic equations, transport equations in spaces of measures, and renormalized/weak solutions for PDEs with singular data. Their diverse formulations and solution concepts enable rigorous analysis of systems with discontinuities, impulses, singularities, and mean-field interactions.

1. Core Definitions and Frameworks

MDEs are formulated in several mathematically rigorous settings:

a. Scalar and vectorial measure-driven equations:

Given a Borel measure μ on [a, b], a scalar MDE is

$dx(t) = f(t, x(t))\,\mu(dt)$

with integral form

$x(t) = x(a) + \int_a^t f(s, x(s))\,\mu(ds)$

A vectorial MDE in ℝⁿ generalizes to

$y(t) = y_0 + \int_{t_0}^t f(s, y(s))\,dg(s)$

where g is a nondecreasing left-continuous function inducing suitable Lebesgue–Stieltjes measures (Pouso et al., 2018).

b. Probability measure evolution (Piccoli–type):

The evolution of a probability measure μ_t ∈ 𝒫(ℝⁿ) is driven by a probability vector field (PVF) V: $$\dot\mu_t = V[\mu_t],\qquad (\pi_1)_\# V[\mu] = \mu$$ with the weak form

$$\frac{d}{dt} \int f(x) \, d\mu_t(x) = \int_{T\R^n} \nabla f(x)\cdot v \, dV[\mu_t](x,v)$$

(Piccoli, 2017, Camilli et al., 2019, Garavello et al., 31 Jan 2026).

c. Generalized ODEs (GODEs) via the Kurzweil–Stieltjes integral:

A generalized ODE is

$\frac{dx}{d\tau}=D[K(x,t)]$

where K may be Kurzweil–Stieltjes integrable with respect to a function of bounded variation. The solution x is regulated and may be discontinuous; dx/dτ is a formal notation rather than a classical derivative (Lu et al., 2023, Weijie et al., 3 Jun 2025, Gallegos et al., 2022).

d. Renormalized solutions to integro-differential equations:

In linear equations with measure data

$- A u = \mu$

where A is a self-adjoint L²-generator, solutions are defined in the sense of renormalization, involving truncations and decomposition of μ into diffuse and concentrated parts (Klimsiak, 2020).

2. Existence, Uniqueness, and Well-posedness

Existence and uniqueness results in the MDE literature depend on both the analytic structure of the problem and the regularity of the driving measures or vector fields:

• Existence: Classical Carathéodory conditions on f and local integrability of the measure ensure the existence of regulated solutions to scalar and vectorial MDEs (Pouso et al., 2018, Gallegos et al., 2022). In the Piccoli framework, if the PVF V is sublinear and narrowly continuous (or locally Lipschitz in an appropriate Wasserstein-type fiber distance), uniform-in-time existence of measure-valued solutions is guaranteed; convergence is via Lattice Approximate Solution (LAS) schemes (Piccoli, 2017, Camilli et al., 2019, Garavello et al., 31 Jan 2026).
• Uniqueness: For PVFs, uniqueness of weak solutions or a semigroup holds only if the barycenter velocity field wμ is globally Lipschitz in both x and μ; otherwise, nonuniqueness and solution selection phenomena arise (Camilli et al., 2019). For linear problems with Dirichlet forms and general measures, uniqueness and m-a.e. identification with integral (Green potential) solutions is established via renormalized solution theory; equivalence with duality solutions is proved (Klimsiak, 2020).
• Peano and Picard-type theorems: The measure control framework parallels classical ODE theory: continuity/compactness and Lipschitz continuity of vector fields in the space of probability measures under Wasserstein and fiber metrics yield existence and well-posedness (Garavello et al., 31 Jan 2026).
• Nonlinear and impulsive problems: In the generalized Kurzweil–Stieltjes framework, conditions such as exponential dichotomy in the linear part and small Lipschitz constants of the nonlinear part are required for unique, bounded, globally defined solutions (Lu et al., 2023, Weijie et al., 3 Jun 2025).

3. Structural Properties, Solution Techniques, and Numerical Schemes

a. Solution spaces and regularity:

MDE solutions may be regulated (i.e. possessing left and right limits everywhere but not necessarily continuous), of bounded variation, or probability measure curves continuous in the narrow topology or Wasserstein metric.

b. Existence and construction (lower/upper solutions):

For (vectorial) MDEs with quasimonotone nondecreasing nonlinearities, the method of lower/upper solutions applies, and minimal/maximal extremal solutions are characterized through monotone operators, appropriate compactness, and regularity (Pouso et al., 2018).

c. Renormalized and integral solutions:

For equations driven by singular (possibly fractal/concentrated) measures, solutions are constructed by truncation and approximation, yielding convergence in narrow topologies and characterizations in terms of Dirichlet energy or duality (Klimsiak, 2020).

d. Numerical schemes:

Discretization methods such as the LAS scheme, semi-discrete Lagrangian methods, and mean-velocity (barycenter) schemes allow numerical evolution of MDEs in spaces of measures. Convergence and error estimates in Wasserstein-type distances are available, although uniqueness may fail for non-Lipschitz cases (Camilli et al., 2019).

e. Superposition principle:

Generalized Ambrosio–Gigli–Savaré superposition results relate measure evolutions to path-space (trajectory) measures concentrated on solutions of associated ODEs, enabling existence and selection mechanisms for solutions (Camilli et al., 2019).

4. Model Classes and Applications

MDEs model a wide variety of mathematical and physical phenomena:

a. Kinetic and transport theory:

MDEs capture the evolution of probability densities under diffusion, finite-speed transport, and concentration effects; the kinetic/mean-field limit of large particle systems is naturally described in this setting (Piccoli, 2017, Camilli et al., 2019).

b. Control theory and uncertainty:

Measure controls generalize relaxed controls; the evolution of measure-valued states under possibly randomized or uncertain controls is framed as an MDE, with classical ODEs, deterministic flows, and stochastic systems recovered as special cases (Garavello et al., 31 Jan 2026).

c. Renormalized PDEs and Dirichlet forms:

Linear (and semilinear) integro-differential equations with general measure data, including fractional Laplacians and divergence-form operators, fit in the MDE/renormalized framework. This is indispensable in treating singular sources and non-local/non-smooth phenomena (Klimsiak, 2020).

d. Impulsive, hybrid, and Stieltjes systems:

By choosing the measure driving the integral appropriately (e.g., having jump discontinuities or time-scale structure), MDEs encode impulsive systems, dynamic equations on time scales, and hybrid dynamics (Gallegos et al., 2022, Lu et al., 2023, Pouso et al., 2018).

e. Nonlinear filtering:

Moment differential equations (MDEs) and derivative-free formulations play a fundamental role in continuous-discrete nonlinear stochastic filtering, particularly for derivative-free EKF variants, where the covariance evolution is described via MDEs or associated sample-point equations (Kulikova et al., 2024).

f. Periodic and bifurcation analysis:

Bifurcation phenomena for periodic solutions of nonlinear MDEs (with impulses and distributional derivatives) can be analyzed via operator-theoretic/topological degree tools in Kurzweil-integral settings. Necessary and sufficient conditions are established for the emergence of non-trivial periodic solution branches (Stefani et al., 2024).

5. Advanced Theoretical Developments

a. Stable and invariant manifolds:

Generalized stable manifold theorems have been extended to nonlinear MDEs under exponential dichotomy hypotheses. Construction relies on generalized Lyapunov–Perron equations and fixed-point arguments in Kurzweil-integral frameworks; the resulting invariant manifolds are typically only Lipschitz-regular in the measure-driven context (Weijie et al., 3 Jun 2025).

b. Linearization and topological conjugacy:

Hartman–Grobman-type theorems for MDEs (generalized ODEs) establish topological conjugacy between linear and nonlinear measure-driven systems, with the conjugacy mappings being proven Hölder continuous under appropriate dichotomy and regularity constraints (Lu et al., 2023).

c. Hybrid and periodic solutions:

Existence results encompass regulated, continuous, differentiable, and S-asymptotically ω-periodic solutions, with the qualitative robustness of solutions under parameter variations analyzed using D-contractions and fixed-point approaches tailored for Banach spaces of regulated functions (Gallegos et al., 2022).

6. Links to Other Differential Equations and Mean-Field Limits

MDEs generalize and unify classical ODEs, PDEs, impulsive, hybrid, and stochastic systems:

• Relation to continuity and transport equations:

When the probability vector field V is induced by a deterministic vector field or state-dependent kernel, the MDE recovers the nonlocal continuity equation or classical transport PDE (Piccoli, 2017, Camilli et al., 2019).

• Stieltjes and generalized derivatives:

Every Stieltjes differential equation (D_g x = F(t,x)) is equivalent to an MDE driven by the measure dg, and the theory applies verbatim to analysis, extremal solutions, and regularity (Pouso et al., 2018, Gallegos et al., 2022).

• Mean-field interaction limits:

Multi-particle dynamical systems converge, in the mean-field limit, to the unique solution of an MDE, extending Dobrushin's classic results to measure-dependent systems (Piccoli, 2017, Camilli et al., 2019).

7. Research Directions and Open Problems

• Selection principles and nonuniqueness:

In general, the set of weak solutions to an MDE may be too large for uniqueness, but additional regularity (fiber-wise Lipschitz, contractivity) restores uniqueness. Analytic and probabilistic selection principles tied to numerical approximation or additional structures remain an active area (Camilli et al., 2019, Garavello et al., 31 Jan 2026).

• Applications to singular PDEs and hybrid systems:

Ongoing work addresses further equivalence and regularity properties of solutions under highly singular measures (purely atomic, fractal, concentrated on null-capacity sets), the extension of bifurcation and stability theory in the Kurzweil–Stieltjes integral setting, and the analytic treatment of hybrid and time-scale-driven systems (Klimsiak, 2020, Stefani et al., 2024, Gallegos et al., 2022).

• Algorithmic and computational development:

Efficient and robust numerical schemes—LAS, sample-point, square-root, hybrid approaches—continue to be refined, with stability, error control, and selection mechanisms at the forefront (Kulikova et al., 2024, Camilli et al., 2019).

---

Key literature and advances:

(Piccoli, 2017, Camilli et al., 2019, Garavello et al., 31 Jan 2026, Klimsiak, 2020, Lu et al., 2023, Weijie et al., 3 Jun 2025, Gallegos et al., 2022, Pouso et al., 2018, Kulikova et al., 2024, Stefani et al., 2024).