Partial Conservation Laws: Core Concepts

• Partial Conservation Laws (PCLs) are identities holding only on specific invariant subspaces, providing conservation properties in restricted contexts.
• They are derived through symmetry reduction techniques in PDEs, stochastic control, and piecewise geometries, yielding constants of motion along invariant solutions.
• PCLs enhance analytical and computational approaches in mathematical physics and control theory by capturing conservation insights where global laws do not apply.

Partial conservation laws (PCLs) extend the standard notion of conservation laws to scenarios where invariance or conservation holds on a restricted subset of states, solutions, or spacetime regions. The term encompasses rigorous methodologies in the theory of partial differential equations, stochastic control, and the mathematical physics of piecewise-defined geometries, enabling one to extract conserved or invariance-type properties outside the standard globally-conserved setting.

1. Foundational Notions and Definitions

Partial conservation laws represent identities or invariance relations that hold only under specific circumstances: for solutions invariant under a symmetry subgroup, for certain sub-manifolds or domains in spacetime, or for functionals of stochastic processes subject to intervention or reset dynamics. Formally, a PCL typically takes the form of a conservation/divergence-type identity that is satisfied solely on certain invariant subspaces, solution manifolds, or across glued regions, rather than globally.

For systems of partial differential equations (PDEs), a classical local conservation law is a divergence identity of the form

$D_t(T) + D_x(X) = 0$

that holds for all solutions. In the PCL setting, such an identity is replaced by conservation holding only for symmetry-invariant solutions or along specific trajectories, invoking more refined algebraic and analytic structures (Druzhkov et al., 2024, Zhang, 2014).

In stochastic models, PCLs appear as marginal conservation relations connecting increments in expected reward and work under thresholds or policies, materially impacting indexability and optimal control (Niño-Mora et al., 11 Jan 2026).

Piecewise geometrical constructions in general relativity employ PCLs via distributional divergences, quantifying conservation up to boundary or shell terms at glued hypersurfaces (Dray, 2017).

2. PCLs in Symmetry Reduction of PDEs

The construction of PCLs for symmetry-invariant solutions of PDE systems is algorithmic and relies on the interplay between symmetries and conservation laws in extended Kovalevskaya form. Given a local symmetry generator $Y=\xi^t\partial_t+\xi^x\partial_x+\eta\partial_u$, its evolutionary characteristic is $Q$, and the symmetry-invariant surface is defined by $Q=0$ together with its differential consequences.

If a local conservation law $(T,X)$ is $Y$-invariant, then the Lie derivative $\mathcal{L}_X\omega$ (with $\omega=T\,dx-X\,dt$) is exact up to a total derivative. On the invariant surface $Q=0$, one derives a reduced conservation law

$D_s(\hat\Psi(\sigma, [u_{(k)}])) = 0$

where $\sigma$ is the group-invariant variable and $\hat\Psi$ is explicit in reduced coordinates. This asserts $\hat\Psi$ is constant along symmetry-invariant solutions, yielding constants of motion for the ODEs governing invariant reduction (Druzhkov et al., 2024).

The reduced constants, called partial conservation laws, essentially serve as first integrals for the reduced ODEs, dramatically facilitating exact solutions and qualitative analysis.

3. Nonlinear Self-Adjointness and Adjoint Symmetries

The adjoint symmetry framework leads to a constructive approach for PCLs via nonlinear self-adjointness: any adjoint symmetry $v(x, u, u_{(1)},\ldots)$ of a system $E_\alpha=0$ is a differential substitution for nonlinear self-adjointness, and vice versa (Zhang, 2014).

The main theoretical result is a formula that builds conserved currents from any symmetry characteristic $Q$ and any adjoint symmetry $\Phi$:

$$D_i\,C^i = Q^\alpha\,E^*_\alpha(x, u, \Phi, \ldots)$$

where $E^*_\alpha$ is the adjoint system, and $C^i$ is computed through repeated integration by parts of the formal Lagrangian $\mathcal{L} = v^\alpha E_\alpha$. Conservation holds only on solutions, thus defining a PCL in this generalized context.

Multipliers, i.e., functions generating conservation laws, are precisely adjoint symmetries satisfying strict variational constraints. The inclusion

$$\text{Multipliers} \subset \text{Adjoint symmetries}$$

follows directly (Zhang, 2014). This method unifies Noether’s theorem and direct multiplier constructions.

4. Distributional and Piecewise Conservation Laws in Geometry

Piecewise conservation laws arise in spacetimes composed by gluing manifolds along non-null hypersurfaces. Let $J^\mu$ be a current which is conserved in each region except possibly at the interface $\Sigma$. Decomposing $J^\mu$ with Heaviside functions:

$$J^\mu(x) = (1 - \Theta(x))\,J^-{}^\mu(x) + \Theta(x)\,J^+{}^\mu(x)$$

and using the distributional calculus, the divergence is

$\nabla_\mu J^\mu = [J^\mu]\,n_\mu\,\delta(\Sigma),$

where $[J^\mu]$ denotes the jump across $\Sigma$, and $n_\mu$ is the normal 1-form (Dray, 2017).

Integrating over a region $V$ crossing $\Sigma$ yields the boundary term as the sole nontrivial contribution:

$$\int_V \nabla_\mu J^\mu\,dV = \int_\Sigma [J^\mu]\,n_\mu\,d\Sigma.$$

This formalism underpins critical results in general relativity, e.g., junction conditions (Israel’s condition), Komar energy jumps, and signature-changing manifolds.

5. PCLs in Restless Bandit Models and Stochastic Control

PCLs are central to restless multi-armed bandit models wherein belief-state processes are subject to reset-type intervention dynamics. For the single-patient problem in treatment adherence outreach, two expectation functionals are defined:

$$F(x,\pi) = \mathbb{E}_{x}^\pi\left[\sum_{t \ge 0} r(X_t, A_t)\,\beta^t\right], \quad G(x,\pi) = \mathbb{E}_{x}^\pi\left[\sum_{t \ge 0} A_t\,\beta^t\right]$$

for reward and work metrics under policy $\pi$. Under threshold policies, the marginal productivity index is

$m(x,z) = \frac{f(x,z)}{g(x,z)}$

where $f(x,z)$ and $g(x,z)$ are differences in $F$ and $G$ between active and passive interventions.

Verifying the three PCLI conditions—strict positivity, monotonicity, and the conservation-type integral relation—ensures Whittle indexability. The third condition, a partial conservation law,

$$F(x,z_2) - F(x,z_1) = \int_{(z_1,z_2]} m(u)\,G(x,du)$$

establishes a Lebesgue-Stieltjes integral structure for reward increments. The optimal cut-off threshold for each intervention price $\lambda$ is given analytically by inverting $m(x)$, and the associated Lagrangian value is piecewise-affine and convex (Niño-Mora et al., 11 Jan 2026).

6. Algorithmic and Computational Aspects

PCL construction in symmetric PDE reduction proceeds algorithmically:

• Compute the symmetry characteristics $Q$ via determining equations.
• Find adjoint symmetries (differential substitutions) via the adjoint system.
• Use conservation law formulae to build partial conservation identities.
• For threshold policies in stochastic control, determine explicit forms for $F$, $G$, and $m(x)$, verify PCLI conditions, invert $m(x)$ for thresholds, and compute dual bounds efficiently (Druzhkov et al., 2024, Niño-Mora et al., 11 Jan 2026).

Implementation in symbolic systems such as Maple leverages total derivative operators, evolutionary fields, and algorithmic computation of invariants and constants of motion.

7. Context, Interpretation, and Significance

PCLs generalize classical conservation law theory to enable extraction of constants of motion, construction of exact group-invariant solutions, and efficient dual bounding in control-theoretic models. In geometric settings, they rigorously account for non-conserved fluxes at interfaces or shell-like discontinuities, essential in gravitational theory and mixed-signature manifolds.

PCLs do not provide full conservation in the usual sense but capture conservation law structure restricted to subsets of interest—whether symmetry-invariant solutions, subdomains, Markovian regimes, or across geometric junctions. This specification is crucial for qualitative and computational advances that would otherwise be inaccessible via global conservation laws alone.

A plausible implication is that further research on generalized and nonlocal PCLs, robust numerical algorithms, and extensions to stochastic and non-variational settings will continue to refine their utility in mathematical physics, control, and applied mathematics.