{"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","summary":"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…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-01","reactions":{}}