(4.21), Remark 4.12, p. 439 — d_H(u, u′) = ‖log u − log u′‖_var
ProvedCompOT.Sinkhorn.eq_4_21hilbert-projective-metricoptimal-transportp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-bookp2o-v1
Let and let have positive entries. Hilbert's projective metric equals the variation seminorm of the difference of the logarithms:
where is applied entrywise.
By this logarithmic change of variables the Hilbert metric on rays of the positive cone becomes a seminorm that vanishes exactly on constant vectors; it turns rates in for Sinkhorn's scalings into rates for the dual potentials .
Preamble
import Mathlib import Definitions.Def_CompOT_Sinkhorn_Defs
Formal statement
namespace CompOT.Sinkhorn
/-- (4.21), Remark 4.12, p. 439: after a logarithmic change of variables, Hilbert's
projective metric is the variation seminorm,
`d_H(u, u') = ‖log u − log u'‖_var`. -/
theorem eq_4_21 {n : ℕ} (hn : 0 < n) (u u' : Fin n → ℝ)
(hu : ∀ i, 0 < u i) (hu' : ∀ i, 0 < u' i) :
hilbertMetric u u' = varNorm (fun i => Real.log (u i) - Real.log (u' i)) := by sorry
end CompOT.Sinkhorn
Source
Peyré & Cuturi, Computational Optimal Transport (FnT ML 2019), Remark 4.12, (4.21), p. 439
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.