Nonlinear Complementarity Problem (NCP) Formulation

• Nonlinear Complementarity Problem (NCP) is defined by conditions x ≥ 0, F(x) ≥ 0, and xᵀF(x) = 0, offering a versatile framework in nonlinear analysis.
• The NCP formulation generalizes the linear complementarity problem and incorporates tensor and polynomial extensions to model complex equilibrium systems.
• Specialized algorithms, including semismooth Newton and homotopy continuation methods, provide robust computational solutions under specific mapping properties.

The nonlinear complementarity problem (NCP) is a core formulation in mathematical programming and nonlinear analysis, underlying equilibrium models in economics, engineering, and beyond. An NCP seeks a vector $x \in \mathbb{R}^n$ such that $x \geq 0$, $F(x) \geq 0$, and $x^T F(x) = 0$ for a given mapping $F : \mathbb{R}^n \to \mathbb{R}^n$. This structure generalizes the classical linear complementarity problem (LCP) and supports a vast array of modeling frameworks, including tensor-structured systems, polynomial map complements, and equilibrium models with nonconvex constraints. Rigorous conditions on $F$, such as monotonicity, $P$-function properties, or specific tensor classes, determine existence, uniqueness, and compactness of solutions, and drive the development of specialized algorithmic schemes for their solution.

1. Mathematical Structure of the NCP

Fundamentally, the NCP is defined by seeking $x \in \mathbb{R}^n$ with the components satisfying

$$x_i \geq 0, \quad F_i(x) \geq 0, \quad x_i F_i(x) = 0, \quad i=1,\dots,n,$$

or, in vector notation,

$x \geq 0, \quad F(x) \geq 0, \quad x^T F(x) = 0.$

This set of conditions encodes so-called complementarity: for each $i$, either $x_i = 0$, $F_i(x) = 0$, or both. In application, $F(x)$ may arise as the derivative of an objective subject to inequality constraints, a nonlinear network flow law, or as the principal part of a higher-order polynomial or tensor mapping (Haddou et al., 2010, Che et al., 2015).

When $F(x)$ is linear, the NCP reduces to the LCP. In advanced settings, $F$ may be polynomial, tensor-based, or defined implicitly by equilibrium models or partial differential equations. Reformulations via NCP-functions, like $\varphi(a,b)=0$ characterizing $a\ge0,\ b\ge0,\ ab=0$, allow replacement by nonlinear system equations amenable to Newton-type or homotopy algorithms (Chen et al., 2015, Dutta et al., 2022).

2. Existence and Uniqueness Theorems

Well-posedness of the NCP is fundamentally tied to properties of the map $F$ and the algebraic structure underlying it. Key criteria include:

• $P_0$-functions: $F$ is called a $P_0$-function if for all $x \neq y$,

$$\max_{i:\,x_i \neq y_i}\;(x_i-y_i)\,(F_i(x)-F_i(y)) \ge 0,$$

ensuring nonnegative principal minors of certain Jacobian matrices (Haddou et al., 2010, Osmani et al., 2022).

• Monotonicity & Generalized Monotonicity: $F$ is monotone when $(x-y)^T (F(x) - F(y)) \ge 0$ for all $x, y$. Strengthened versions are required for error estimation and global convergence in smoothing and homotopy methods (Haddou et al., 2010, 1207.1145).
• Structured Tensors: For tensor-based NCPs, positivity and copositivity are crucial. A symmetric, positive definite tensor $A$ of even order ($m\ge2$) yields

$$A x^m > 0 \quad \forall x \in \mathbb{R}^n\setminus\{0\}, \qquad A x^m = \sum_{i_1,\dots,i_m} A_{i_1\dots i_m}x_{i_1}\cdots x_{i_m},$$

and copositivity is defined on $\mathbb{R}_+^n$ (Che et al., 2015). If $A$ is diagonalizable and positive definite, the NCP $(q, A)$ has a unique solution; if merely positive definite or strictly copositive, the solution set is nonempty and compact.

• Degree Theory for Polynomials and Tensors: For polynomial $F$, the boundedness and existence of solutions are guaranteed if the residual map $\Phi(x)=\min\{x, f^*(x)\}$ has degree $1$ and only trivial solution at $0$ (Gowda, 2016). R–tensor constructions ensure the "strong Q–property": global existence for arbitrary perturbations of the leading term.

3. Reformulation Techniques: Smoothing and Homotopy

To enable numerical solution, NCPs are frequently reformulated using smoothing functions or homotopy maps:

• NCP–functions and Smoothers: Functions $\varphi(a,b)$ are constructed to satisfy $\varphi(a,b)=0 \Leftrightarrow a\ge0, b\ge0, ab=0$. Common examples include:
  • Fischer–Burmeister: $\varphi_{FB}(a,b)=a+b-\sqrt{a^2+b^2}$
  • Min–function: $\varphi_{min}(a, b) = \min(a, b)$
  • Smoothing families: $\theta(t)$ and $\psi(t)=1-\theta(t)$, with rational and exponential examples (Haddou et al., 2010, Osmani et al., 2022, Daniilidis et al., 2024).
• Smoothing System Embedding: Replace original complementarity conditions by smooth equations such as $G_r(s, t) = r\,\psi^{-1}[\psi(s/r)+\psi(t/r)] = 0$, with $r \downarrow 0$ (Haddou et al., 2010, Osmani et al., 2022). Algorithmic frameworks leverage Newton-type methods with variable $r$, line-search, and Armijo criteria.
• Homotopy Continuation: Path-following procedures embed the NCP into a higher-dimensional system with a homotopy parameter $\lambda \in [0,1]$,

$$H(x, x^{(0)}, \lambda) = (1-\lambda)\,\text{auxiliary equations} + \lambda\,\text{original system},$$

and trace the solution from known initial to the target system as $\lambda \to 0$ (1207.1145, Dutta et al., 2022, Dutta et al., 2022, Dutta et al., 2022).

4. Tensor, Polynomial, and Structured NCPs

NCPs frequently arise in settings where $F$ is of higher algebraic structure:

• Tensor NCPs: For $A\in T_{m,n}$, $F(x) = A x^{m-1} + q$, and solution theory exploits tensor symmetries, positive definiteness, and diagonalizability (Che et al., 2015, Dutta et al., 2022).
• Polynomial NCPs (PCP): Where $F$ is polynomial, the solution set SOL($f$, $q$) can be characterized via the leading homogeneous term $f^*$, with compactness and existence controlled by degree-theoretic results (Gowda, 2016). Matrix constructions produce tensors inheriting complementarity properties from classical matrices under specific powers.
• Eigenvalue Complementarity: NCPs govern eigenvalue problems for tensors, with special semismooth reformulation and Newton-type solution methods (Chen et al., 2015).
• Traffic Equilibrium as NCP: Finsler geometry provides an alternative, equipping $\mathbb{R}^n$ with position- and direction-dependent metrics for encoding equilibrium criteria and gap functions (Asanjarani, 2021).

5. Algorithmic Solution and Convergence Theory

Efficient algorithms for NCPs rely on a range of principles:

• Semi-smooth Newton: For strongly semismooth NCP-functions, Newton-type schemes are globally convergent under $P_0$ or monotonicity assumptions, with locally superlinear or quadratic rates (Chen et al., 2015, Osmani et al., 2022, Bui et al., 2018).
• Homotopy Path–Following: Homotopy maps guarantee bounded, non-intersecting solution paths under weak nonsingularity or monotonicity, with probability-one global convergence in typical Newton–Fixed–Point Homotopy methods (1207.1145).
• Regularization for NAVE: Nonlinear absolute value equations can be reformulated as NCPs via variable-splitting and solved by smoothing-regularization, exploiting $P_0$ properties for robust convergence (Daniilidis et al., 2024).
• Composite NCP/Homotopy Systems: Slack variables and composite blockwise reformulations embed complementarity in smooth auxiliary systems, supporting bounded predictor–corrector arc-length methods (Dutta et al., 2022, Dutta et al., 2022).

6. Applications and Model-Specific Formulations

NCPs have pervasive importance across domains:

• Compositional Multiphase Flow: NCP reformulation avoids ill-defined primary variables in phase transitions, facilitating robust, scalable Newton-type solution even for large-scale heterogeneous benchmarks (Bui et al., 2018).
• Oligopoly/Cournot Equilibrium: Market optimization models employ NCPs for Nash equilibrium computation, where homotopy-path continuation ensures convergence from arbitrary starting points (Dutta et al., 2022, Dutta et al., 2022).
• Transit Assignment under Priority Rules: Schedule-based transit equilibrium requires arc-level NCP formulations to capture FCFS and continuance priority. MPEC reformulations using Fischer–Burmeister merit functions provide tractable algorithms for realistic networks and capture all behavioral constraints inherent in group- and arc-level priority enforcement (Feng et al., 12 Jan 2026).
• Traffic Equilibrium in Finsler Spaces: Finsler geometrical programming yields dynamical NCP formulations, enabling geodesic-based minimization and new interpretations of classical equilibrium flows (Asanjarani, 2021).

7. Extensions, Limitations, and Directions

The core NCP framework admits generalization to systems of nonlinear equations via NCP-function equivalence, higher-order polynomial and tensor complementarity, and smoothing or semismooth algorithms robust under weak regularity conditions. Existence and uniqueness theorems depend crucially on tensor and map properties, and degree theory provides a universal language for analyzing global solvability (Gowda, 2016). Some smoothing schemes (e.g., strong semismoothness required in TEiCP) exclude fully classical methods, necessitating tailored algorithmic approaches (Chen et al., 2015).

A plausible implication is that as NCP formulations generalize to more complex algebraic and networked systems, algorithmic flexibility, careful structural analysis (e.g., copositivity, diagonalizability, monotonicity, $P_0$-functions), and tailored reformulation techniques become essential for both theoretical guarantees and practical computational tractability.

---