AFTD Auto-Formalizing Theoretical Domains

op223_separated_branch_derivative_positive

theoremverified

For finite nonnegative vectors p and q, suppose 0 ≤ r < 1, q is pointwise at least r times p, and equality holds at u, v, and j. If p at u exceeds p at v and p at j is positive, then the product of q's u-v difference and its j entry, divided by one plus its squared norm, is strictly smaller than the analogous quantity for p.

Statement

theorem op223_separated_branch_derivative_positive {ι : Type*} [Fintype ι] (p q : ι → ℝ) (u v j : ι) (r : ℝ) (hr0 : 0 ≤ r) (hr1 : r < 1) (hp : ∀ i, 0 ≤ p i) (hq : ∀ i, 0 ≤ q i) (hpj : 0 < p j) (huv : p v < p u) (hbound : ∀ i, r * p i ≤ q i) (hqu : q u = r * p u) (hqv : q v = r * p v) (hqj : q j = r * p j) : (q u - q v) * q j / (1 + ∑ i, q i ^ 2) < (p u - p v) * p j / (1 + ∑ i, p i ^ 2)
Source
arXiv:2610.17078 Open Problem 6.1; scalar inequality for a separated-branch derivative argument
Verified
12 Nov 2026
Axioms
Classical.choiceQuot.soundpropext

The proof, explained

The proof clears the positive denominators to reduce the statement to a cross-multiplied product inequality. Because q matches r * p at indices u, v, and j, the numerator on the left equals r ^ 2 times the numerator on the right, which is strictly positive since p u > p v and p j > 0. Meanwhile, the pointwise bound r * p i ≤ q i together with r ^ 2 < 1 guarantees that r ^ 2 times the denominator of the right quotient is strictly smaller than the denominator of the left quotient. Multiplying these two inequalities completes the proof.

  1. The inequality between the two quotients reduces to the cross-multiplied inequality (q u - q v) * q j * (1 + ∑ i, p i ^ 2) < (p u - p v) * p j * (1 + ∑ i, q i ^ 2).

    Both denominators 1 + ∑ i, p i ^ 2 and 1 + ∑ i, q i ^ 2 are strictly positive because each is 1 plus a sum of squares of real numbers (established by `positivity`). Clearing denominators via `div_lt_div_iff₀` therefore reduces the comparison of quotients to the cross-multiplied inequality. Line 4 merely introduces an unused local hypothesis.

    lines 4–7

    uses div_lt_div_iff₀

      have _ := hq
      have hDp : 0 < 1 + ∑ i, p i ^ 2 := by positivity
      have hDq : 0 < 1 + ∑ i, q i ^ 2 := by positivity
      rw [div_lt_div_iff₀ hDq hDp]
  2. r ^ 2 * (1 + ∑ i, p i ^ 2) < 1 + ∑ i, q i ^ 2.

    From 0 ≤ r < 1, it follows that r ^ 2 < 1. For each index i, since r ≥ 0 and p i ≥ 0, their product is nonnegative (`mul_nonneg`), so squaring the bound r * p i ≤ q i preserves the inequality. Distributing the exponent via `mul_pow` gives r ^ 2 * p i ^ 2 ≤ q i ^ 2. Distributing r ^ 2 inside the summation with `Finset.mul_sum` and summing these termwise inequalities over all i using `Finset.sum_le_sum` shows that r ^ 2 * ∑ i, p i ^ 2 ≤ ∑ i, q i ^ 2. Adding r ^ 2 < 1 to this sum inequality yields r ^ 2 * (1 + ∑ i, p i ^ 2) < 1 + ∑ i, q i ^ 2 by linear arithmetic (`linarith`).

    lines 8–17

    uses Finset.mul_sum, Finset.sum_le_sum, mul_nonneg, mul_pow

      have hr2 : r ^ 2 < 1 := by nlinarith
      have hsum : r ^ 2 * ∑ i, p i ^ 2 ≤ ∑ i, q i ^ 2 := by
        rw [Finset.mul_sum]
        apply Finset.sum_le_sum
        intro i _
        have h1 : 0 ≤ r * p i := mul_nonneg hr0 (hp i)
        have h2 : (r * p i) ^ 2 ≤ q i ^ 2 := by nlinarith [hbound i]
        have h3 : (r * p i) ^ 2 = r ^ 2 * p i ^ 2 := mul_pow r (p i) 2
        linarith
      have hD : r ^ 2 * (1 + ∑ i, p i ^ 2) < 1 + ∑ i, q i ^ 2 := by linarith
  3. 0 < (p u - p v) * p j.

    Because p v < p u, the difference p u - p v is strictly positive (`sub_pos.mpr`). Since p j > 0 as well, their product (p u - p v) * p j is strictly positive (`positivity`).

    lines 18–20

    uses sub_pos.mpr

      have hN : 0 < (p u - p v) * p j := by
        have : 0 < p u - p v := sub_pos.mpr huv
        positivity
  4. (q u - q v) * q j = r ^ 2 * ((p u - p v) * p j).

    Substituting the equalities q u = r * p u, q v = r * p v, and q j = r * p j turns the expression into (r * p u - r * p v) * (r * p j). Factoring out r from the difference and collecting powers gives r ^ 2 * ((p u - p v) * p j), which is verified by polynomial identity expansion (`ring`).

    lines 21–23

      have hnum : (q u - q v) * q j = r ^ 2 * ((p u - p v) * p j) := by
        rw [hqu, hqv, hqj]
        ring
  5. The cross-multiplied inequality holds, completing the proof.

    Rewriting the left numerator as r ^ 2 * ((p u - p v) * p j), the goal becomes r ^ 2 * ((p u - p v) * p j) * (1 + ∑ i, p i ^ 2) < (p u - p v) * p j * (1 + ∑ i, q i ^ 2). Since (p u - p v) * p j > 0 and r ^ 2 * (1 + ∑ i, p i ^ 2) < 1 + ∑ i, q i ^ 2, this inequality follows directly by non-linear arithmetic (`nlinarith`).

    lines 24–25

      rw [hnum]
      nlinarith
div_lt_div_iff₀
For positive denominators b and d, the inequality a / b < c / d is equivalent to a * d < c * b.
Finset.mul_sum
Multiplication distributes over finite sums: a * ∑ i ∈ s, f i = ∑ i ∈ s, a * f i.
Finset.sum_le_sum
If f(i) ≤ g(i) for all i in a finite set s, then ∑ i ∈ s, f i ≤ ∑ i ∈ s, g i.
mul_nonneg
The product of two nonnegative numbers is nonnegative.
mul_pow
Powers distribute over multiplication in a commutative monoid: (a * b) ^ n = a ^ n * b ^ n.
sub_pos.mpr
If b < a, then 0 < a - b.

Written by a model (gemini-3.8-flash) from the Lean proof and the goal states Lean reported at each step. Lean verified the proof; this explanation is a reading aid and is not itself machine-checked.

Lean source view module on GitHub

open scoped Classicaltheorem op223_separated_branch_derivative_positive {ι : Type*} [Fintype ι] (p q : ι → ℝ) (u v j : ι) (r : ℝ) (hr0 : 0 ≤ r) (hr1 : r < 1) (hp : ∀ i, 0 ≤ p i) (hq : ∀ i, 0 ≤ q i) (hpj : 0 < p j) (huv : p v < p u) (hbound : ∀ i, r * p i ≤ q i) (hqu : q u = r * p u) (hqv : q v = r * p v) (hqj : q j = r * p j) : (q u - q v) * q j / (1 + ∑ i, q i ^ 2) < (p u - p v) * p j / (1 + ∑ i, p i ^ 2) := by  have _ := hq  have hDp : 0 < 1 + ∑ i, p i ^ 2 := by positivity  have hDq : 0 < 1 + ∑ i, q i ^ 2 := by positivity  rw [div_lt_div_iff₀ hDq hDp]  have hr2 : r ^ 2 < 1 := by nlinarith  have hsum : r ^ 2 * ∑ i, p i ^ 2 ≤ ∑ i, q i ^ 2 := by    rw [Finset.mul_sum]    apply Finset.sum_le_sum    intro i _    have h1 : 0 ≤ r * p i := mul_nonneg hr0 (hp i)    have h2 : (r * p i) ^ 2 ≤ q i ^ 2 := by nlinarith [hbound i]    have h3 : (r * p i) ^ 2 = r ^ 2 * p i ^ 2 := mul_pow r (p i) 2    linarith  have hD : r ^ 2 * (1 + ∑ i, p i ^ 2) < 1 + ∑ i, q i ^ 2 := by linarith  have hN : 0 < (p u - p v) * p j := by    have : 0 < p u - p v := sub_pos.mpr huv    positivity  have hnum : (q u - q v) * q j = r ^ 2 * ((p u - p v) * p j) := by    rw [hqu, hqv, hqj]    ring  rw [hnum]  nlinarith
See it in the graphDiscuss this resultChallenge it