Binary tilted mean
DefinitionTiltedCvx_Supplement_binMeanconvex-analysisgeneralized-convexitynonsmooth-analysisoptimizationp2o-gran-per-chapterp2o-plan-bookp2o-v1
For , and ,
and for , (15). The mean multiplies by the weight of the smaller value; is the arithmetic mean . For it is the upper expectile at level of the two-point distribution. It gives the Jensen form of tilted secant convexity (Theorem 18).
Formalization Note The convention is a separate branch; for , the denominators are positive.
Definition code
import Mathlib
namespace TiltedCvx.Supplement
/-- **The binary mean `M_{γ,λ}(a, b)` ((14)–(15)).** Pedramfar–Khanna–Quinn–Aggarwal,
arXiv:2611.03472, (14) and (15), p.25: for `γ ∈ (0, 1]`, `λ ∈ [0, 1]` and `a, b ∈ ℝ`,
`M_{γ,λ}(a, b) := (γλa + (1 − λ)b)/(γλ + (1 − λ))` if `a ≤ b`, and
`(λa + γ(1 − λ)b)/(λ + γ(1 − λ))` if `a > b`; for `γ = 0`, `M_{0,λ}(a, b) := max{a, b}`.
**Formalization Note** The `γ = 0` convention (15) is a separate branch. For `γ ∈ (0, 1]` and
`λ ∈ [0, 1]` both denominators are positive; only these parameters are used. -/
noncomputable def binMean (γ l a b : ℝ) : ℝ :=
if γ = 0 then max a b
else if a ≤ b then (γ * l * a + (1 - l) * b) / (γ * l + (1 - l))
else (l * a + γ * (1 - l) * b) / (l + γ * (1 - l))
end TiltedCvx.Supplement
Source
Pedramfar–Khanna–Quinn–Aggarwal, Tilted Convexity and an Exact Offline–Online Separation, arXiv:2611.03472, (14)–(15), p.25