One-Sided Lipschitz Conditions

• One-Sided Lipschitz conditions are generalized monotonicity constraints that bound the quadratic form of increments rather than their full norm.
• They enable uniqueness, stability, and convergence in complex systems such as differential equations, stochastic dynamics, and control problems with stiff or superlinear nonlinearities.
• Analytical characterization via the symmetric part of the Jacobian guides computational methods, error control, and practical numerical approximation schemes.

A one-sided Lipschitz condition is a generalized monotonicity-type constraint imposed on nonlinear maps, vector fields, or set-valued mappings, that relaxes the classical global Lipschitz (bi-Lipschitz) requirement by bounding the quadratic form associated with the increment instead of its norm. This relaxation plays a central role in contemporary analysis, control, stochastic dynamics, differential inclusions, and numerical approximation—especially for systems admitting superlinear or "stiff" nonlinearities. One-sided Lipschitz conditions enable uniqueness, stability, and convergence claims in situations where two-sided Lipschitz continuity is either inapplicable or unnecessarily conservative. The literature recognizes variants such as local, global, relaxed, and negative one-sided Lipschitz constants, each tailored to specific applications in PDEs, SDEs, control theory, and variational analysis.

1. Formal Definitions and Analytical Framework

The canonical one-sided Lipschitz (OSL) condition for a function $f: \mathbb{R}^n \to \mathbb{R}^n$ is: $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$ with $\lambda \in \mathbb{R}$ the OSL constant. This bounds the symmetric part of the directional derivative, and $\lambda$ can be negative for contractive or monotone flows. In variational, PDE, and SPDE applications, the OSL property is often encoded via Gelfand triples or variational pairings, e.g., for a Nemytskii operator in a Hilbert space,

$2\,_{V^*}\!\langle \Phi(x)-\Phi(y),\,x-y\rangle_{V} \leq L\|x-y\|^2_{\mathcal{H}}$

for all $x, y$ in the Banach or Hilbert framework (Beyn et al., 2017, Sauer et al., 2013).

Relaxed (set-valued) OSL, as for multivalued maps $F: [0,T] \times V \to 2^H$ with closed, convex values, reads

$$\langle g-\hat g,\,u-v\rangle_{V^*,V} \leq \ell(t) \|u-v\|_H^2,\quad g\in F(t,u),\;\exists\,\hat g\in F(t,v)$$

for some measurable rate $\ell\in L^1$ (Beyn et al., 2017, Eberhard et al., 2019).

In the context of time-dependent velocity fields (e.g., transport or SDE flows), the OSL inequality appears as

$(b(t,x) - b(t,y))\cdot(x-y) \geq -C_1(t) |x-y|^2$

with $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$0 and $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$1 possibly only measurable in $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$2 (Camilli et al., 13 May 2025, Lions et al., 2023).

2. Comparison with Classical Lipschitz and Monotonicity Hypotheses

A globally Lipschitz map $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$3 satisfies $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$4, which in turn implies the directional OSL condition with $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$5 by Cauchy–Schwarz. However, OSL is strictly weaker: it restricts only the component of $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$6 along $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$7, not the full norm disparity. One-sided Lipschitz functions need not be (two-sided) Lipschitz, nor even locally so. Prototypical examples include $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$8 and $$\langle f(x) - f(y),\,x-y \rangle \leq \lambda\,\|x-y\|^2,\quad\forall\,x, y \in \mathcal{D}$$9, both globally OSL with $\lambda \in \mathbb{R}$0, but not globally nor locally Lipschitz. Dissipativity—i.e., OSL with $\lambda \in \mathbb{R}$1—is particularly significant for ensuring contractivity of the dynamical flow, exponential stability, and mixing in SDEs (Abbaszadeh et al., 2013, Bréhier, 2020, Majka, 2016).

In multivalued and differential inclusion settings, the OSL condition enables the existence and uniqueness of trajectories (solutions), even under irregular, possibly discontinuous nonlinearities—something full Lipschitz continuity would exclude (Beyn et al., 2017, Eberhard et al., 2019).

3. Computation and Analytical Characterization of OSL Constants

For $\lambda \in \mathbb{R}$2 maps, the minimal OSL constant $\lambda \in \mathbb{R}$3 over a compact set $\lambda \in \mathbb{R}$4 is characterized by the symmetric part of the Jacobian,

$\lambda \in \mathbb{R}$5

This formulation underpins computational techniques—grid sampling, interval analysis, branch-and-bound, and SOS relaxations—to derive sharp or conservative $\lambda \in \mathbb{R}$6 in practice (Nugroho et al., 2020, Coënt et al., 2019, Coënt et al., 2019).

For reaction–diffusion and switched control systems, spatial/temporal discretizations yield symmetric matrices (e.g., discrete Laplacians), so OSL analysis often reduces to spectral radius computations. In infinite dimensions or for operator-valued nonlinearities, the pairing is taken in the energy or dual norm (Sauer et al., 2013, Beyn et al., 2017).

4. Applications in Differential Equations, Stochastic Analysis, and Numerical Schemes

Semilinear Parabolic Inclusions and PDEs

Relaxed one-sided Lipschitz conditions facilitate well-posedness (existence of weak solutions) and compactness for parabolic differential inclusions, even when nonlinearities grow superlinearly. The key is the ability to dominate the nonlinearity in energy estimates, yielding a Gronwall-type control and compactness in solution trajectories under Galerkin discretization (Beyn et al., 2017).

SDEs with Superlinear Drift

Local and global OSL drifts (possibly with only local OSL growth) enable existence and uniqueness results for SDEs, as in the classic work of Krylov for locally monotone coefficients (Tan et al., 2018, Majka, 2016). For numerical approximation, OSL conditions justify implicit and tamed explicit schemes—crucial for systems with superlinear drifts, e.g., $\lambda \in \mathbb{R}$7 (Tan et al., 2018, Bréhier, 2020). The OSL structure controls divergence in moments and ensures strong convergence (in $\lambda \in \mathbb{R}$8 and in probability) of the scheme interpolants.

Transport, Fokker–Planck, and Conservation Laws

In linear and nonlinear transport equations with only one-sided Lipschitz velocity fields, OSL guarantees existence and stability of (Filippov) flows, unique viscosity solutions for backward/forward problems, and strong regular Lagrangian solutions in the expansive regime. The OSL constant prescribes the maximal exponential separation between characteristics and directly bounds the compressibility of the stochastic flow Jacobian in SDEs and related Fokker–Planck equations (Lions et al., 2023, Camilli et al., 13 May 2025).

Control, Observer and Verification

In nonlinear observer design, the OSL constant enters Lyapunov-based or LMI stability criteria, generally yielding less conservative conditions than classical Lipschitz bounds—critical for high-gain observers of stiff, monotone, or highly nonlinear systems (Abbaszadeh et al., 2013, Nugroho et al., 2020). For switched or sampled systems, OSL constants facilitate sharp truncation and error accumulation bounds in semi-Lagrangian grid-based controller synthesis and optimal reachability, enabling guaranteed safety and performance verification over finite and infinite horizons (Coënt et al., 2019, Coënt et al., 2019).

5. One-Sided Lipschitz in Numerical Approximation and Convergence

OSL constants appear explicitly in quantifying the strong convergence rate and error propagation in finite-difference, finite-element, and grid-based numerical schemes for PDEs/SPDEs. For reaction-diffusion SPDEs, the OSL (monotonicity) bound on the drift enables absorption of high-degree nonlinear growth in error estimates and ensures convergence at explicit rates, with constants depending exponentially on the OSL parameter (Sauer et al., 2013). In the scalar conservation law and follow-the-leader approximation, discrete forms of the OSL condition are equivalent to discrete entropy admissibility, ensuring convergence to entropy solutions and preservation of the admissibility condition at the discrete level (Francesco et al., 2021).

For SDE and Fokker–Planck evolutions, the OSL property ensures that numerical schemes (explicit Euler, tamed Euler, implicit θ-methods) remain stable, with error bounds and moment bounds growing at most polynomially in time, proportional to the OSL constant (Bréhier, 2020, Tan et al., 2018, Coënt et al., 2019, Camilli et al., 13 May 2025).

In grid-based feedback synthesis for control or safety, OSL-based error propagation is central to bounding the discrepancy between continuous and discrete (Euler or SDE-driven) system evolution (Coënt et al., 2019, Coënt et al., 2019).

6. Multivalued and Relaxed OSL: Variational and Set-Valued Analysis

Set-valued mappings that are L-relaxed one-sided Lipschitz with $\lambda \in \mathbb{R}$9 admit explicit localization and preimage structure stronger than mere metric regularity, permitting precise geometric characterization of solution sets via intersections and unions of balls centered on boundary points of the image sets. This property is critical in variational analysis and in the study of differential inclusions or sweeping processes, where full Lipschitz parameterizations are unavailable (Eberhard et al., 2019, Beyn et al., 2017).

The main preimage formula for such $\lambda$0 is: $\lambda$1 and this can be further reduced to extreme points of $\lambda$2 for computational tractability (Eberhard et al., 2019).

7. Illustrative Models and Examples

• Semilinear PDE with Nemytskii Nonlinearity: $\lambda$3 and set-valued thickenings yield one-sided Lipschitz nonlinearity of arbitrary polynomial order, admitting existence/convergence of Galerkin approximations (Beyn et al., 2017).
• Superlinear Drift SDE: $\lambda$4 is OSL but not Lipschitz, handled by truncated (tamed) schemes (Tan et al., 2018).
• FitzHugh–Nagumo SPDE: Nonlinear drift $\lambda$5 is OSL, not Lipschitz—finite-difference approximations converge strongly with explicit rate dependence on the OSL constant (Sauer et al., 2013).
• Scalar Conservation Law: The follow-the-leader scheme imposes a discrete OSL (entropy) condition at the particle level, ensuring convergence to the entropy solution (Francesco et al., 2021).
• Explicit Preimage Calculation: For $\lambda$6, $\lambda$7, the L-ROSL formula exactly recovers the classical inverse, illustrating sharp localization beyond metric regularity (Eberhard et al., 2019).

8. Theoretical and Practical Impact

One-sided Lipschitz conditions have become a foundational analytical tool in modern nonlinear science, allowing treatment of large classes of dissipative or monotone systems, differential inclusions, and SDEs with superlinear drift, for which two-sided Lipschitz constraints are inaccessible. They underpin the feasibility and optimality of weak solution constructions, observer and controller design, ergodicity and mixing in stochastic systems, and robust numerical approximation for high-dimensional and high-nonlinearity settings.

The flexibility and comparative sharpness of OSL bounds continue to drive methodological advances in mathematics, engineering, and scientific computing, often yielding less conservative estimates, tractable synthesis conditions, and effective algorithms in contexts where full Lipschitz structure is either impossible or intractable.