[Paper Review] A tropical extremal problem with nonlinear objective function and linear inequality constraints

Nikolai Krivulin|arXiv (Cornell University)|Dec 26, 2012

Polynomial and algebraic computation17 references3 citations

TL;DR

This paper presents a novel closed-form solution for a multidimensional tropical extremal problem with a nonlinear objective function and linear inequality constraints over an idempotent semifield. By reformulating the problem as a linear inequality in an extended variable space, the authors derive an exact solution using spectral radius and trace-based identities, validated through a two-dimensional numerical example in the $\mathbb{R}_{\max,+}$ semifield.

ABSTRACT

We consider a multidimensional extremal problem formulated in terms of tropical mathematics. The problem is to minimize a nonlinear objective function, which is defined on a finite-dimensional semimodule over an idempotent semifield, subject to linear inequality constraints. An efficient solution approach is developed which reduces the problem to that of solving a linear inequality with an extended set of unknown variables. We use the approach to obtain a complete solution to the problem in a closed form under quite general assumptions. To illustrate the obtained results, a two-dimensional problem is examined and its numerical solution is given.

Motivation & Objective

• To develop a complete, closed-form solution for a multidimensional extremal problem with a nonlinear objective function in tropical mathematics.
• To address the gap in existing literature by providing a direct, non-iterative solution under general assumptions, unlike prior methods that yield only partial or iterative results.
• To extend unconstrained tropical extremal solutions to constrained settings with linear inequality constraints.
• To establish a systematic approach based on matrix trace identities and spectral radius computation for solving such problems.
• To validate the method through a detailed two-dimensional numerical example in the $\mathbb{R}_{\max,+}$ semifield.

Proposed method

• Reformulate the original nonlinear tropical extremal problem into an equivalent linear inequality system with an extended set of unknown variables.
• Utilize a new binomial identity for matrix traces in the context of idempotent semifields to derive necessary conditions for optimality.
• Apply spectral radius and trace-based expressions to characterize the optimal value of the objective function.
• Use the star-closure operation $(\cdot)^*$ on matrices to represent the solution set of the resulting linear inequality system.
• Derive the solution as a parametric family of vectors generated by the columns of the star-closure of the matrix $\theta^{-1}A \oplus B$, where $\theta$ is the optimal value.
• Ensure the solution satisfies both the nonlinear objective minimization and all linear inequality constraints through algebraic equivalence.

Experimental results

Research questions

• RQ1 Can a complete closed-form solution be derived for a tropical extremal problem with a nonlinear objective function and linear inequality constraints?
• RQ2 How can the problem be transformed into an equivalent linear inequality system to enable direct solution?
• RQ3 What role do matrix trace and spectral radius play in characterizing the optimal value in such problems?
• RQ4 Under what general conditions does the proposed method yield a complete solution set?
• RQ5 How can the solution be explicitly constructed and verified in a finite-dimensional tropical setting?

Key findings

• The optimal value $\theta$ is given by the maximum over $k=1$ to $n$ of the $k$-th root of the trace of products of matrices $AB^{i_1}\cdots AB^{i_k}$, plus the spectral radius of $A$.
• The solution set is expressed as $\bm{x} = (\theta^{-1}A \oplus B)^* \bm{u}$ for any regular vector $\bm{u}$, with the star-closure operation ensuring feasibility.
• In the two-dimensional example, the optimal value is $\theta = 2$, and the solution lies along a ray generated by the vector $\bm{x}_0 = \begin{pmatrix} 0 \\ 5 \end{pmatrix}$.
• The method successfully handles both the nonlinear objective and linear constraints simultaneously, yielding a complete solution where previous approaches only offered partial or iterative results.
• The derived solution is consistent with known results for unconstrained problems when $B = \mathbb{0}$, confirming theoretical validity.
• The approach is general and applies to any radicable, linearly ordered idempotent semifield, with the $\mathbb{R}_{\max,+}$ semifield as a key illustrative case.