Abstract:In this article, we present new general results on existence of augmented Lagrange multipliers. We define a penalty function associated with an augmented Lagrangian, and prove that, under a certain growth assumption on the augmenting function, an augmented Lagrange multiplier exists if and only if this penalty function is exact. We also develop a new general approach to the study of augmented Lagrange multipliers called the localization principle. The localization principle allows one to study the local behaviour of the augmented Lagrangian near globally optimal solutions of the initial optimization problem in order to prove the existence of augmented Lagrange multipliers.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is about the existence problem of augmented Lagrange multipliers. Specifically, the author proposes two new general methods to study the existence of augmented Lagrange multipliers: 1. **Exact Penalty Function Method**: - The author defines a penalty function related to the augmented Lagrangian function and proves that under certain growth conditions, the augmented Lagrange multiplier exists if and only if this penalty function is exact. This method allows the use of the existing exact penalty function theory to prove the existence of the augmented Lagrange multiplier. - In particular, the author proves that for the sharp Lagrangian, the augmented Lagrange multiplier exists if and only if the standard \( \ell_1 \) penalty function is exact. 2. **Localization Principle**: - The localization principle allows the study of the local behavior of the augmented Lagrangian near the global optimal solution of the original optimization problem, thereby proving the existence of the augmented Lagrange multiplier. - By using sufficient optimality conditions, the local analysis can be simplified to the study of optimality conditions, thus transforming the existence problem of the augmented Lagrange multiplier into the study of optimality conditions. ### Main Contributions - **Exact Penalty Function Method**: By introducing a penalty function related to the augmented Lagrangian function and proving the equivalence between its exactness and the existence of the augmented Lagrange multiplier, a new proof method is provided. - **Localization Principle**: A new method is proposed to prove the existence of the augmented Lagrange multiplier by studying the local behavior of the augmented Lagrangian near the global optimal solution. ### Applications - These methods can be applied to various optimization problems such as mathematical programming problems and nonlinear semidefinite programming problems, providing new tools and perspectives for the theoretical study of the augmented Lagrangian method. ### Formulas - The augmented Lagrangian function \( L(x, \lambda, r) \) is defined as: \[ L(x, \lambda, r) = \inf_{p \in P} \left( \Phi(x, p) - \langle \lambda, p \rangle + r \sigma(p) \right) \] - The penalty function \( F(x, r, C) \) is defined as: \[ F(x, r, C) = \inf_{p \in C} \left( \Phi(x, p) + r \sigma(p) \right) \] - The definition of the augmented Lagrange multiplier: \[ v(p) \geq v(0) + \langle \lambda, p \rangle - r \sigma(p) \quad \forall p \in P \] Through these methods and definitions, the paper provides new solutions to the existence problem of augmented Lagrange multipliers.

Existence of augmented Lagrange multipliers: reduction to exact penalty functions and localization principle

Exact Augmented Lagrangian Duality for Mixed Integer Convex Optimization

A New Penalty Dual-Primal Augmented Lagrangian Method and Its Extensions

Exact penalty functions with multidimensional penalty parameter and adaptive penalty updates

Second-Order Conditions for the Existence of Augmented Lagrange Multipliers for Sparse Optimization

On Saddle Points of Augmented Lagrangians for Constrained Nonconvex Optimization

A New Insight on Augmented Lagrangian Method with Applications in Machine Learning

A nonsmooth exact penalty method for equality-constrained optimization: complexity and implementation

Exact augmented Lagrangians for constrained optimization problems in Hilbert spaces II: Applications

A unifying theory of exactness of linear penalty functions II: parametric penalty functions

A simple proof of existence of Lagrange multipliers

Study on Initial Value of Lagrange Multipliers and Adaptation of Penalty Parameters

Implicit augmented Lagrangian and generalized optimization

A class of smooth exact penalty function methods for optimization problems with orthogonality constraints

Local convergence of augmented Lagrangian methods for composite optimization

An Augmented Lagrangian Approach to Composite Problems with a Random Linear Operator

The inexact power augmented Lagrangian method for constrained nonconvex optimization

Smooth exact penalty functions II: a reduction to standard exact penalty functions

Global Convergence Analysis of the Power Proximal Point and Augmented Lagrangian Method

Convergence analysis of primal-dual augmented Lagrangian methods and duality theory

Morozov's principle for the augmented Lagrangian method applied to linear inverse problems