Telescoping:
ProvedPersonalizedAssortment.IB.telescoping_sumLet be any function and a natural number. Let be natural numbers with that fall by or in each step: for . Then
Each unit sold is counted once, at the inventory level just before its sale. This is the pathwise content of the final equality in the paper's computation of the expected dual objective.
Formalization Note The path is indexed from . The left side is written with real-valued differences, and the right side sums over the integers with .
import Mathlib
namespace PersonalizedAssortment.IB
/-- The final equality of the expected-dual display in §A.3 (p. 1549), pathwise: if the inventory
of a product starts at `c` and falls by `0` or `1` in each of `T` periods, then
`∑_{t=1}^T Ψ(I^{t-1}/c)(I^{t-1} - I^t) = ∑_{k=I^T+1}^{c} Ψ(k/c)` (the `k`th sold unit
contributes `Ψ((c - k + 1)/c)`). Periods are indexed from `0` here. -/
theorem telescoping_sum (Ψ : ℝ → ℝ) (c : ℕ) (T : ℕ) (I : ℕ → ℕ) (hI0 : I 0 = c)
(hstep : ∀ t < T, I (t + 1) ≤ I t ∧ I t ≤ I (t + 1) + 1) :
∑ t ∈ Finset.range T, Ψ ((I t : ℝ) / (c : ℝ)) * ((I t : ℝ) - (I (t + 1) : ℝ)) =
∑ k ∈ Finset.Ioc (I T) c, Ψ ((k : ℝ) / (c : ℝ)) := by sorry
end PersonalizedAssortment.IB
Read-back
What the Lean code literally says, in plain math · claude-opus-5-5
Let be an arbitrary function; nothing at all is assumed about it. Let and be natural numbers. Let be a sequence of natural numbers such that:
- ;
- for every , , so each of the first steps decreases the value by or .
Values for are unconstrained and do not appear. The statement asserts
The differences on the left are taken in the reals. On the right, runs over the integers in .
Degenerate cases.
- If , both sides are : the left is an empty sum, and the right runs over the empty range .
- If , the hypotheses force . Every ratio uses , and both sides are .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.