Id: arxiv/2610.30144 Url: https://arxiv.org/abs/2610.30144 Tags: math.OC Score: 0 Title: Forward Tangent-Centered Sinkhorn DRO: Exact Nominal Recovery from Optimal Transport to KL Domain: math.OC Source: arxiv Authors: Junjie Guo Reviews: 0 Subject: math.OC Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-08-01 Page: https://delphi.science/r/arxiv/2610.30144 Fixed-reference entropy-regularized transport yields smooth distributionally robust optimization (DRO) objectives, but its value as a function of the adversarial marginal is generally minimized at a kernel-smoothed law rather than at the nominal distribution. Consequently, the smallest raw ambiguity set need not recover nominal risk. Building on prior information-geometric plan-KL orientations, we study the forward tangent-centered discrepancy obtained by subtracting the value and first variation at the nominal law. This is the unique affine recentering, up to the simplex gauge; for…