Monotone Numerical Schemes

Updated 14 February 2026

Monotone numerical schemes are discretization methods that enforce discrete maximum and comparison principles to guarantee stability and convergence for nonlinear or degenerate equations.
They employ strategies like upwind/wide-stencil discretizations and M-matrix constructions to control oscillations and achieve robust numerical performance.
Applications span Hamilton–Jacobi–Bellman equations, conservation laws, nonlocal operators, and fractional PDEs, with extensions to high-dimensional and path-dependent problems.

Monotone numerical schemes are a class of discretization methods for differential and integral equations that enforce discrete versions of comparison, maximum, or minimum principles. Monotonicity is a critical structural property that guards against spurious oscillations, ensures stability, and allows for convergence proofs to generalized (“weak” or “viscosity”) solutions in highly nonlinear or degenerate settings. These schemes have played a central role in the analysis and simulation of nonlinear PDEs, stochastic control problems, conservation laws, nonlocal operators, and fractional and path-dependent equations.

1. Structural Definition and Foundational Principles

A discrete operator $\mathcal{T}_h$ is monotone if, whenever two grid functions $u\le v$ (componentwise), it holds that $\mathcal{T}_h[u]\le \mathcal{T}_h[v]$ . For schemes formulated via backward/implicit time stepping, this translates to each update being nondecreasing in its solution arguments. For explicit, upwind-type, or finite-volume schemes, monotonicity arises from the numerical flux $F(a,b)$ being nondecreasing in $a$ and nonincreasing in $b$ , or, in the case of general nonlinear schemes, from the matrix structure being an $M$ -matrix at each step (nonpositive off-diagonals and maximal row diagonal dominance).

Monotonicity is a sufficient (though not always necessary) condition for enforcing discrete maximum principles and comparison results. Its application allows the extension of the Barles–Souganidis framework, which establishes convergence to viscosity solutions for scalar, nonlinear, degenerate elliptic or parabolic PDEs provided the scheme is consistent, monotone, and stable (Ren et al., 2015, Guo et al., 2012, Debrabant et al., 2014).

2. Monotone Schemes for Viscosity Solutions of PDEs

Many fully nonlinear PDEs lack smooth solutions and must be interpreted in the viscosity framework. For a broad class of equations—including Hamilton–Jacobi–Bellman (HJB), Isaacs, and Monge–Ampère—the comparision principle for viscosity solutions hinges on the monotonicity of the associated schemes.

Finite-Difference and Semi-Lagrangian Schemes: Consistency and monotonicity are typically enforced via upwind or wide-stencil discretizations and positive, convex combinations of values at stencil points. In high dimensions or for fully nonlinear, degenerate elliptic equations, this often necessitates wide stencils and monotone interpolation (Debrabant et al., 2014, Chen et al., 2016, Guo et al., 2012). For the Monge–Ampère equation, mixed schemes alternate between second-order narrow and monotone wide stencils to retain both accuracy and monotonicity (Chen et al., 2016).
Nonlocal and Fractional Operators: For nonlocal Isaacs or fractional Laplace-type equations, monotone quadrature or piecewise-linear collocation with nonnegative weights is used. Regularization and mesh grading techniques enable monotonicity and yield pointwise and free-boundary convergence rates (Borthagaray et al., 2024, Biswas et al., 2017).
Path-Dependent PDEs: Schemes for PPDEs are constructed using discrete nonlinear expectation operators satisfying a monotonicity condition, allowing convergence results for non-Markovian problems, stochastic differential games, and second-order BSDEs (Ren et al., 2015).

3. Monotonicity in Conservation Laws and Parabolic Systems

In scalar conservation laws, monotone upwind finite-volume schemes are defined so that their update is nondecreasing in each argument, or, equivalently, their flux is monotone. Classical Godunov, Engquist–Osher, and Lax–Friedrichs fluxes are monotone in this sense (Fjordholm et al., 2016, Badwaik et al., 2019).

Convergence and Stability: Monotone schemes guarantee $L^1$ contraction and uniqueness for the weak (entropy) solution. For strictly monotone fluxes, sharp rates of $O(\sqrt{\Delta x})$ can be achieved for discontinuous flux (Badwaik et al., 2019). For certain initial data, monotone schemes deliver $O(\Delta x^2)$ convergence in Wasserstein distance, despite only first-order $L^1$ accuracy (Fjordholm et al., 2016).
Block Monotone Iterative Schemes: Coupled systems with quasi-monotone reaction terms admit block Jacobi and block Gauss–Seidel approaches where the monotone property underpins convergence and the discrete maximum principle. The existence and uniqueness of solutions is achieved via upper/lower sequence iteration within an ordered sector (Al-Sultani, 2019).

4. Monotonicity in Nonlocal, Fractional, and Time-Fractional Equations

Monotonicity in this context ensures the fundamental discrete comparison and positivity properties crucial for nonlocal or memory equations.

Fractional ODEs and Subdiffusion: A scheme is called complete monotonicity-preserving (CM-preserving) if the coefficients in the discrete convolution (after inversion) form a CM sequence (i.e., all forward differences are nonnegative). CM-preserving schemes are $A(\pi/2)$ -stable, admit discrete comparison principles, and preserve monotonicity for scalar nonlinear autonomous fractional ODEs; the L1 scheme is of this class (Li et al., 2019).
Nonlocal Integrodifferential Operators: Two-scale monotone discretizations combine near-field kernel regularization and mesh collocation to ensure all weights are nonnegative, establishing a discrete comparison principle. Mesh grading and regularization scales are adjusted to optimize convergence and stability (Borthagaray et al., 2024).

5. High-Dimensional and Probabilistic Monotone Approaches

For high-dimensional fully nonlinear equations, monotone schemes employ stochastic representations (trinomial trees, Monte Carlo sampling) with carefully chosen weights and moments so all update coefficients remain nonnegative. This circumvents the curse of dimensionality inherent in deterministic mesh-based stencils and extends monotonicity-based convergence to dimensions $d\gtrsim 12$ (Guo et al., 2012). Such schemes rely on the Barles–Souganidis framework and exploit diagonal-only Hessian dependence, with explicit parameter regimes guaranteeing nonnegativity and permitting convergence analysis.

6. Monotonicity in Galerkin and Finite Element Frameworks

Monotonicity in discontinuous Galerkin (dG) methods is achieved by augmenting the base discretization with artificial diffusion derived from shock detectors and graph-Laplacian operators. Nonlinear stabilization is tailored to suppress spurious local extrema while maintaining Lipschitz continuity and the discrete maximum principle. Twice-differentiable smoothing of the nonlinearities enables efficient Newton-type solvers, and the resulting operator guarantees DMP/LED properties, existence, and improved solver efficiency (Badia et al., 2016). In the context of anisotropic diffusion, nonnegative directional splitting decomposes the operator so that the resulting finite difference matrix is an $M$ -matrix, ensuring monotonicity; stencil size is dictated by the anisotropy of the diffusion tensor (Ngo et al., 2015).

7. Applications, Convergence Rates, and Limitations

Monotone numerical schemes are broadly applied to:

Hamilton–Jacobi–Bellman and Isaacs equations in deterministic and stochastic control, including nonlocal and nonconvex variants, where monotonicity yields explicit rates and stability under degenerate or singular operator structure (Biswas et al., 2017, Debrabant et al., 2014, Guo et al., 2012, Borthagaray et al., 2024).
Path-dependent and nonlocal PDEs, including those driven by jump diffusions, fractional covariances, or memory effects.
Nonlinear phase-field equations, where mono-tonicity-in-time is a sharp guarantee of correct steady-state convergence, and energy stability is insufficient (Li et al., 2024).
Robust handling of sharp layers, shocks, and discontinuities in conservation laws, with provably minimal oscillation and maximally efficient convergence rates in appropriate solution metrics (Fjordholm et al., 2016, Badwaik et al., 2019).
Large-scale stochastic variational inequalities and monotone inclusions, where monotonic splitting and variance-reduction techniques yield convergence with near-optimal oracle complexity (Cui et al., 2020).

Limitations include the possible need for wide stencils (increasing resolution cost for strong anisotropy or high order), conservative accuracy in non-smooth or fully nonlinear regimes, and (in the nonlocal case) memory or computation constraints associated with global operators. Advanced mesh grading, local regularization, or adaptive quadrature strategies are employed to mitigate these costs (Borthagaray et al., 2024). In fully nonlinear, nonconvex, or multidimensional settings, monotonicity may restrict the class of usable discretizations or necessitate complexity-reducing approximations (Guo et al., 2012, Debrabant et al., 2014).

References: