Hidden Convexity in Multi-Retailer Cold-Chain Allocation: A Cholesky-Spectral Second-Order Cone Programming Approach
math.OC
Submitted: 2026-10-26
Updated: 2026-10-26
License: http://creativecommons.org/licenses/by/4.0/
The gist: Several dairy retailers share a refrigerated cold-chain network, each allocating products across heat-sensitivity subgroups to maximize quadratic-fractional efficiency (DEA style).
Terminology
Abstract
Several dairy retailers share a refrigerated cold-chain network, each allocating products across heat-sensitivity subgroups to maximize quadratic-fractional efficiency (DEA style). This formulation measures spoilage risk per unit of logistics cost, separating technical from scale efficiency. The single-agent problem is a non-convex fractional QCQP, while the multi-retailer game forms a generalized Nash equilibrium problem (GNEP) with a shared risk-budget constraint, which is generally NP-hard. Governed by the Arrhenius spoilage law, the thermal-risk covariance inherits a rank-one-plus-diagonal structure that unlocks hidden convexity. The authors propose the HSDI (Hidden SOCP Dinkelbach Iteration) algorithm via a joint Cholesky-spectral transform. Key contributions include: 1. Establishing simultaneous diagonalization of without requiring matrix commutativity. 2. Deriving an explicit sufficient condition for exact second-order cone programming (SOCP) relaxation, alongside a closed-form suboptimality bound for inexact regimes. 3. Proving global convergence: the outer Dinkelbach loop converges Q-superlinearly (exact) or R-linearly (inexact), and the coupled variational-equilibrium (VE) iteration converges R-linearly under strong monotonicity, achieving an overall complexity while ensuring Lipschitz stability of the primal-dual pair. Calibrated on Vietnam and US dairy markets, HSDI matches the objective value of multi-start non-linear baselines without random restarts, yielding global optima under exactness. HSDI exhibits an empirical scaling exponent of 0.93 (versus 1.30 for SLSQP), offering a certified polynomial-time framework and a stable shadow price for prioritizing cold-chain investments.
Related papers
- An Accelerated Proximal Gradient Method for Bures-Wasserstein Space
- FOMTools: Development and AI-Assisted Code Generation for First-Order Methods
- Lions and Muons: Optimization via Stochastic Frank-Wolfe under Heavy-Tailed Noise
- Dynamic Service Recommendation with Congestion-Dependent Joining: Near-Optimal and Constant-Factor Approximation Algorithms
- Integrating Learning and Optimization for Open Transport Capacity Pooling under Disruption
- Zeroth-Order Gradient Estimators under Decision-Dependent Noise