AFTD Auto-Formalizing Theoretical Domains

op223_inside_branch_fixed_parameter_gap

theoremverified

At a tied resolvent maximum inside the moved branch, the fixed-spectral-parameter derivative of the root-minus-competitor gap is positive under the displayed Schur-complement and positive-definiteness inequalities. This scalar result does not control the derivative of the spectral radius.

Statement

theorem op223_inside_branch_fixed_parameter_gap (t d a b C k f : ℝ)
    (ht0 : 0 ≤ t) (ht1 : t < 1) (hC : 0 < C)
    (hd : 0 < d) (ha : 0 < a) (hda : a < d)
    (hdet : a ^ 2 < d * b) (hCauchy : f ^ 2 ≤ C * k)
    (htie : d = k + f ^ 2 * (t * (t*d+(1-t)*a)+(1-t)*(t*a+(1-t)*b))) :
    C * (d-a) * (t*d+(1-t)*a) -
      f ^ 2 * ((t*d+(1-t)*a) - (t*a+(1-t)*b)) *
        (1+C*(t*(t*d+(1-t)*a)+(1-t)*(t*a+(1-t)*b))) > 0
Source
Scalar reduction for the weighted homotopy in Open Problem 6.1, https://arxiv.org/abs/2610.17078.
Verified
15 Nov 2026
Axioms
Classical.choiceQuot.soundpropext

The proof, explained

The proof establishes that the derivative expression $C(d-a)p - f^2(p-q)(1+Cs)$ is strictly positive by splitting on the sign of $p - q$, where $p = td + (1-t)a$, $q = ta + (1-t)b$, and $s = tp + (1-t)q$. When $p \ge q$, the expression is rewritten via an algebraic identity into $C(1-t)(db-a^2) + (p-q)(Ck-f^2)$, which is strictly positive because the Schur-complement condition $a^2 < db$ makes the first term positive and Cauchy-Schwarz $f^2 \le Ck$ makes the second term non-negative. When $p < q$, the expression is viewed as the difference between the strictly positive term $C(d-a)p$ and the non-positive term $f^2(p-q)(1+Cs)$, which is again strictly positive.

  1. With the convex combinations $p = t d + (1 - t) a$, $q = t a + (1 - t) b$, and $s = t p + (1 - t) q$, the quantity $p$ is strictly positive.

    We define $p$ as the convex combination of $d$ and $a$, $q$ as the convex combination of $a$ and $b$, and $s$ as the convex combination of $p$ and $q$. Since $t \ge 0$ and $d > 0$, the product $t d$ is non-negative (by `mul_nonneg` and `le_of_lt`). Since $t < 1$ and $a > 0$, the term $(1 - t) a$ is strictly positive (by `mul_pos`). Their sum $p$ is therefore strictly positive.

    lines 11–18

    uses mul_nonneg, le_of_lt, mul_pos

      let p := t*d+(1-t)*a
      let q := t*a+(1-t)*b
      let s := t*p+(1-t)*q
      have hp : 0 < p := by
        dsimp [p]
        have h1 : 0 ≤ t*d := mul_nonneg ht0 (le_of_lt hd)
        have h2 : 0 < (1-t)*a := mul_pos (by linarith) ha
        linarith
  2. The tie hypothesis simplifies to $d = k + f^2 s$, and the goal becomes $C (d - a) p - f^2 (p - q) (1 + C s) > 0$.

    Recognizing that the nested convex combination in the tie condition is precisely $s$, we rewrite the tie assumption as $d = k + f^2 s$. The target inequality is then rewritten in terms of $p, q$, and $s$ as $0 < C (d - a) p - f^2 (p - q) (1 + C s)$.

    lines 19–20

      have hs : d = k + f ^ 2 * s := htie
      change 0 < C * (d-a) * p - f ^ 2 * (p-q) * (1+C*s)
  3. In the case $p - q \ge 0$, the expression $C (d - a) p - f^2 (p - q) (1 + C s)$ is strictly positive.

    Assuming $p - q \ge 0$, we decompose the expression into $C (1 - t) (d b - a^2) + (p - q) (C k - f^2)$. The first term is strictly positive because $C > 0$, $1 - t > 0$, and the Schur-complement condition gives $d b - a^2 > 0$. The second term is non-negative by `mul_nonneg` because $p - q \ge 0$ and the Cauchy-Schwarz bound $f^2 \le C k$ gives $C k - f^2 \ge 0$. The algebraic identity equating the expression to this sum is established by expanding via `ring`, substituting $d q - a p = (1 - t) (d b - a^2)$ and $d - f^2 s = k$, and regrouping multiplications using `mul_assoc`. Since the expression equals the sum of a strictly positive term and a non-negative term, it is strictly positive.

    lines 21–35

    uses mul_nonneg, mul_assoc

      by_cases hm : 0 ≤ p-q
      · have h1 : 0 < C * (1-t) * (d*b-a^2) := by positivity
        have h2 : 0 ≤ (p-q) * (C*k-f^2) := mul_nonneg hm (by linarith)
        have hid : C * (d-a) * p - f ^ 2 * (p-q) * (1+C*s) =
            C * (1-t) * (d*b-a^2) + (p-q) * (C*k-f^2) := by
          have hfirst : C * (d-a) * p - f ^ 2 * (p-q) * (1+C*s) =
              C * (d*q-a*p) + (p-q) * (C*(d-f^2*s)-f^2) := by ring
          have hsecond : d*q-a*p = (1-t)*(d*b-a^2) := by
            dsimp [p,q]
            ring
          have hthird : d-f^2*s=k := by linarith [hs]
          rw [hsecond, hthird] at hfirst
          simpa only [mul_assoc] using hfirst
        rw [hid]
        linarith
  4. In the case $p - q < 0$, the expression $C (d - a) p - f^2 (p - q) (1 + C s)$ is strictly positive.

    In the remaining case, $p - q < 0$. The first term $C (d - a) p$ is strictly positive because $C > 0$, $d - a > 0$ (from $a < d$), and $p > 0$. For the second term, the condition $a^2 < d b$ combined with $0 \le a^2$ (from `sq_nonneg` and `lt_of_le_of_lt`) gives $d b > 0$, which implies $b > 0$ by `mul_pos_iff_of_pos_left` since $d > 0$. Consequently, $q = t a + (1 - t) b \ge 0$ and $s = t p + (1 - t) q \ge 0$, ensuring $1 + C s \ge 0$. Since $f^2 \ge 0$, $p - q \le 0$ (from `le_of_lt`), and $1 + C s \ge 0$, their product $f^2 (p - q) (1 + C s)$ is non-positive by `mul_nonpos_of_nonpos_of_nonneg` and `mul_nonpos_of_nonneg_of_nonpos`. Subtracting this non-positive quantity from the strictly positive term $C (d - a) p$ yields a strictly positive result.

    lines 36–49

    uses lt_of_le_of_lt, sq_nonneg, mul_pos_iff_of_pos_left, le_of_lt, mul_nonpos_of_nonpos_of_nonneg, mul_nonpos_of_nonneg_of_nonpos

      · have h1 : 0 < C*(d-a)*p := by positivity
        have h2 : f^2*(p-q)*(1+C*s) ≤ 0 := by
          have hq : 0 ≤ q := by
            dsimp [q]
            have hdb : 0 < d * b := lt_of_le_of_lt (sq_nonneg a) hdet
            have hb : 0 < b := (mul_pos_iff_of_pos_left hd).mp hdb
            positivity
          have hsp : 0 ≤ s := by
            dsimp [s]
            positivity
          have hfactor : 0 ≤ 1+C*s := by positivity
          have hneg : p-q < 0 := by linarith
          exact mul_nonpos_of_nonpos_of_nonneg (mul_nonpos_of_nonneg_of_nonpos (sq_nonneg f) (le_of_lt hneg)) hfactor
        linarith
mul_nonneg
The product of two non-negative numbers is non-negative.
le_of_lt
A strict inequality implies the corresponding non-strict inequality ($a < b \implies a \le b$).
mul_pos
The product of two strictly positive numbers is strictly positive.
mul_assoc
Multiplication is associative: $a * b * c = a * (b * c)$.
lt_of_le_of_lt
If $a \le b$ and $b < c$, then $a < c$.
sq_nonneg
The square of any real number is non-negative ($0 \le a^2$).
mul_pos_iff_of_pos_left
For $a > 0$, $a * b > 0$ if and only if $b > 0$.
mul_nonpos_of_nonpos_of_nonneg
If $a \le 0$ and $b \ge 0$, then $a * b \le 0$.
mul_nonpos_of_nonneg_of_nonpos
If $a \ge 0$ and $b \le 0$, then $a * b \le 0$.

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

set_option maxHeartbeats 800000theorem op223_inside_branch_fixed_parameter_gap (t d a b C k f : ℝ)    (ht0 : 0 ≤ t) (ht1 : t < 1) (hC : 0 < C)    (hd : 0 < d) (ha : 0 < a) (hda : a < d)    (hdet : a ^ 2 < d * b) (hCauchy : f ^ 2 ≤ C * k)    (htie : d = k + f ^ 2 * (t * (t*d+(1-t)*a)+(1-t)*(t*a+(1-t)*b))) :    C * (d-a) * (t*d+(1-t)*a) -      f ^ 2 * ((t*d+(1-t)*a) - (t*a+(1-t)*b)) *        (1+C*(t*(t*d+(1-t)*a)+(1-t)*(t*a+(1-t)*b))) > 0 := by  let p := t*d+(1-t)*a  let q := t*a+(1-t)*b  let s := t*p+(1-t)*q  have hp : 0 < p := by    dsimp [p]    have h1 : 0 ≤ t*d := mul_nonneg ht0 (le_of_lt hd)    have h2 : 0 < (1-t)*a := mul_pos (by linarith) ha    linarith  have hs : d = k + f ^ 2 * s := htie  change 0 < C * (d-a) * p - f ^ 2 * (p-q) * (1+C*s)  by_cases hm : 0 ≤ p-q  · have h1 : 0 < C * (1-t) * (d*b-a^2) := by positivity    have h2 : 0 ≤ (p-q) * (C*k-f^2) := mul_nonneg hm (by linarith)    have hid : C * (d-a) * p - f ^ 2 * (p-q) * (1+C*s) =        C * (1-t) * (d*b-a^2) + (p-q) * (C*k-f^2) := by      have hfirst : C * (d-a) * p - f ^ 2 * (p-q) * (1+C*s) =          C * (d*q-a*p) + (p-q) * (C*(d-f^2*s)-f^2) := by ring      have hsecond : d*q-a*p = (1-t)*(d*b-a^2) := by        dsimp [p,q]        ring      have hthird : d-f^2*s=k := by linarith [hs]      rw [hsecond, hthird] at hfirst      simpa only [mul_assoc] using hfirst    rw [hid]    linarith  · have h1 : 0 < C*(d-a)*p := by positivity    have h2 : f^2*(p-q)*(1+C*s) ≤ 0 := by      have hq : 0 ≤ q := by        dsimp [q]        have hdb : 0 < d * b := lt_of_le_of_lt (sq_nonneg a) hdet        have hb : 0 < b := (mul_pos_iff_of_pos_left hd).mp hdb        positivity      have hsp : 0 ≤ s := by        dsimp [s]        positivity      have hfactor : 0 ≤ 1+C*s := by positivity      have hneg : p-q < 0 := by linarith      exact mul_nonpos_of_nonpos_of_nonneg (mul_nonpos_of_nonneg_of_nonpos (sq_nonneg f) (le_of_lt hneg)) hfactor    linarith
See it in the graphDiscuss this resultChallenge it