op_1693_noon_witness_loss_monotonicity_one
For N = 1 and any balance parameter q in the open interval (0, 1) and transmissivities x, y in the open interval (0, 1), if noon_realign_witness 1 q x y > 1, then the partial derivatives of noon_realign_witness 1 q at x and y are non-negative (0 ≤ deriv (fun u => noon_realign_witness 1 q u y) x and 0 ≤ deriv (fun v => noon_realign_witness 1 q x v) y), and the detection region is upward closed: for 0 < x ≤ x' < 1 and 0 < y ≤ y' < 1, noon_realign_witness 1 q x y > 1 implies noon_realign_witness 1 q x' y' > 1.
Statement
theorem op_1693_noon_witness_loss_monotonicity_one : (∀ (q x y : ℝ) (hq : q ∈ Set.Ioo 0 1) (hx : x ∈ Set.Ioo 0 1) (hy : y ∈ Set.Ioo 0 1), noon_realign_witness 1 q x y > 1 → 0 ≤ deriv (fun u => noon_realign_witness 1 q u y) x ∧ 0 ≤ deriv (fun v => noon_realign_witness 1 q x v) y) ∧ (∀ (q x y x' y' : ℝ) (hq : q ∈ Set.Ioo 0 1) (hx : 0 < x ∧ x ≤ x' ∧ x' < 1) (hy : 0 < y ∧ y ≤ y' ∧ y' < 1), noon_realign_witness 1 q x y > 1 → noon_realign_witness 1 q x' y' > 1)
- Formalizes
- arXiv:2611.17021
- Verified
- 15 Nov 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 2
- noon_realign_witnessFor any natural number N, real balance parameter q, and real transmissivities x, y, the second-moment realignment witness W_N(q, x, y) is…
- Cslib.SKI.Polynomial.evalSubstitute terms for the free variables of a polynomial
Read back from the Lean
The declaration asserts a conjunction of two statements about noon_realign_witness with parameter $N = 1$:
1. For all real numbers $q, x, y$, if $q \in (0, 1)$, $x \in (0, 1)$, and $y \in (0, 1)$, and if $\text{noon\_realign\_witness}(1, q, x, y) > 1$, then:
- the derivative of $u \mapsto \text{noon\_realign\_witness}(1, q, u, y)$ at $x$ is non-negative ($\ge 0$), and
- the derivative of $v \mapsto \text{noon\_realign\_witness}(1, q, x, v)$ at $y$ is non-negative ($\ge 0$).
2. For all real numbers $q, x, y, x', y'$, if $q \in (0, 1)$, $0 < x \le x' < 1$, and $0 < y \le y' < 1$, and if $\text{noon\_realign\_witness}(1, q, x, y) > 1$, then $\text{noon\_realign\_witness}(1, q, x', y') > 1$.
Written by a model that saw only the Lean, never the English above. If the two disagree, that is worth a challenge.
The proof, explained
For photon number N = 1, the realignment witness deficit W_1(q, x, y) - 1 is algebraically proportional to the quadratic form G(s, r) = s^2 + r^2 + 4sr - s - r with s = qx and r = (1-q)y. When W_1 > 1, G(s, r) > 0 forces the linear lower bounds 2s + 4r ≥ 1 and 4s + 2r ≥ 1, which in turn guarantee that the partial derivatives of W_1 with respect to x and y are non-negative. Furthermore, these lower bounds ensure that G(s+ds, r+dr) remains strictly positive under any non-negative increments ds, dr ≥ 0, proving that the detection region is upward closed.
The derivative of any quadratic function u ↦ a*u^2 + b*u + c is 2*a*x + b.
We represent the quadratic function as the evaluation of the polynomial a*X^2 + b*X + c. Differentiating this polynomial expression yields the linear polynomial 2*a*X + b, whose evaluation at x is 2*a*x + b.
uses
Polynomial.derivprivate lemma op_1693_noon_witness_loss_monotonicity_one_deriv_quadratic (a b c x : ℝ) : deriv (fun u => a*u^2 + b*u + c) x = 2*a*x + b := by have h : (fun u => a*u^2 + b*u + c) = fun u => (Polynomial.C a * Polynomial.X^2 + Polynomial.C b * Polynomial.X + Polynomial.C c).eval u := by funext u; simp rw [h] simp only [Polynomial.deriv] simp ring
For N = 1, the witness deficit W_1(q, x, y) - 1 equals 2*((q*x)^2 + ((1-q)*y)^2 + 4*(q*x)*((1-q)*y) - (q*x) - ((1-q)*y)).
Unfolding the definition of noon_realign_witness for N = 1, the sum over range(2) expands into the k=0 and k=1 terms. Simplifying algebraically yields the stated quadratic expression in q*x and (1-q)*y.
uses
Finset.sum_range_succprivate lemma op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one (q x y : ℝ) : noon_realign_witness 1 q x y - 1 = 2*((q*x)^2 + ((1-q)*y)^2 + 4*(q*x)*((1-q)*y) - (q*x) - ((1-q)*y)) := by unfold noon_realign_witness simp [Finset.sum_range_succ, Nat.choose] ring
The partial derivative of the witness with respect to x is 2*q*(2*(q*x) + 4*((1-q)*y) - 1).
Viewing the witness as a quadratic function of the first variable u with coefficients depending on q and y, we apply the quadratic derivative lemma and simplify algebraically.
uses
Finset.sum_range_succ,op_1693_noon_witness_loss_monotonicity_one_deriv_quadraticprivate lemma op_1693_noon_witness_loss_monotonicity_one_witness_one_x_deriv (q y x : ℝ) : deriv (fun u => noon_realign_witness 1 q u y) x = 2*q*(2*(q*x) + 4*((1-q)*y) - 1) := by have hfun : (fun u => noon_realign_witness 1 q u y) = fun u => (2*q^2)*u^2 + (-2*q + 8*q*(1-q)*y)*u + (1 - 2*((1-q)*y) + 2*((1-q)*y)^2) := by funext u unfold noon_realign_witness simp [Finset.sum_range_succ, Nat.choose] ring rw [hfun] rw [op_1693_noon_witness_loss_monotonicity_one_deriv_quadratic] ring
The partial derivative of the witness with respect to y is 2*(1-q)*(4*(q*x) + 2*((1-q)*y) - 1).
Symmetrically, viewing the witness as a quadratic function of the second variable v, applying the quadratic derivative lemma and factoring yields the result.
uses
Finset.sum_range_succ,op_1693_noon_witness_loss_monotonicity_one_deriv_quadraticprivate lemma op_1693_noon_witness_loss_monotonicity_one_witness_one_y_deriv (q x y : ℝ) : deriv (fun v => noon_realign_witness 1 q x v) y = 2*(1-q)*(4*(q*x) + 2*((1-q)*y) - 1) := by have hfun : (fun v => noon_realign_witness 1 q x v) = fun v => (2*(1-q)^2)*v^2 + (-2*(1-q) + 8*(1-q)*q*x)*v + (1 - 2*(q*x) + 2*(q*x)^2) := by funext v unfold noon_realign_witness simp [Finset.sum_range_succ, Nat.choose] ring rw [hfun] rw [op_1693_noon_witness_loss_monotonicity_one_deriv_quadratic] ring
For positive s and r, if s^2 + r^2 + 4*s*r - s - r > 0, then 1 ≤ 2*s + 4*r. automation
Assuming for contradiction that 2*s + 4*r < 1, the expression s^2 + r^2 + 4*s*r - s - r can be rewritten as a sum of squares and non-negative products that is non-positive, contradicting strict positivity.
uses
sq_nonneg,mul_nonnegprivate lemma op_1693_noon_witness_loss_monotonicity_one_G_pos_x_bound (s r : ℝ) (hs : 0 < s) (hr : 0 < r) (hG : 0 < s^2 + r^2 + 4*s*r - s - r) : 1 ≤ 2*s + 4*r := by by_contra h push Not at h nlinarith [sq_nonneg s, sq_nonneg (r - 1), mul_nonneg hs.le (show (0:ℝ) ≤ 1 - 2*s - 4*r by linarith), mul_nonneg hr.le (show (0:ℝ) ≤ 1 - r by linarith)]
For positive s and r, if s^2 + r^2 + 4*s*r - s - r > 0, then 1 ≤ 4*s + 2*r. automation
Symmetrically, assuming 4*s + 2*r < 1 contradicts positivity of the quadratic form via non-linear arithmetic.
uses
sq_nonneg,mul_nonnegprivate lemma op_1693_noon_witness_loss_monotonicity_one_G_pos_y_bound (s r : ℝ) (hs : 0 < s) (hr : 0 < r) (hG : 0 < s^2 + r^2 + 4*s*r - s - r) : 1 ≤ 4*s + 2*r := by by_contra h push Not at h nlinarith [sq_nonneg r, sq_nonneg (s - 1), mul_nonneg hr.le (show (0:ℝ) ≤ 1 - 4*s - 2*r by linarith), mul_nonneg hs.le (show (0:ℝ) ≤ 1 - s by linarith)]
The quadratic form G(s, r) = s^2 + r^2 + 4*s*r - s - r is strictly positive under non-negative shifts ds, dr ≥ 0 when G(s, r) > 0 and s, r > 0.
The bounds 2*s + 4*r - 1 ≥ 0 and 4*s + 2*r - 1 ≥ 0 mean that the first-order variation (2*s + 4*r - 1)*ds + (4*s + 2*r - 1)*dr is non-negative, and the remaining terms ds^2 + dr^2 + 4*ds*dr are also non-negative, so G(s+ds, r+dr) remains strictly positive.
uses
op_1693_noon_witness_loss_monotonicity_one_G_pos_x_bound,op_1693_noon_witness_loss_monotonicity_one_G_pos_y_bound,mul_nonneg,sq_nonnegprivate lemma op_1693_noon_witness_loss_monotonicity_one_G_mono (s r ds dr : ℝ) (hs : 0 < s) (hr : 0 < r) (hds : 0 ≤ ds) (hdr : 0 ≤ dr) (hG : 0 < s^2 + r^2 + 4*s*r - s - r) : 0 < (s+ds)^2 + (r+dr)^2 + 4*(s+ds)*(r+dr) - (s+ds) - (r+dr) := by have hA : 0 ≤ 2*s + 4*r - 1 := by have := op_1693_noon_witness_loss_monotonicity_one_G_pos_x_bound s r hs hr hG; linarith have hB : 0 ≤ 4*s + 2*r - 1 := by have := op_1693_noon_witness_loss_monotonicity_one_G_pos_y_bound s r hs hr hG; linarith nlinarith [mul_nonneg hA hds, mul_nonneg hB hdr, mul_nonneg hds hdr, sq_nonneg ds, sq_nonneg dr]
The theorem splits into two conjuncts: non-negativity of the partial derivatives and monotonicity (upward closure) of the detection region.
We split the conjunction into its two goals.
theorem op_1693_noon_witness_loss_monotonicity_one : (∀ (q x y : ℝ) (hq : q ∈ Set.Ioo 0 1) (hx : x ∈ Set.Ioo 0 1) (hy : y ∈ Set.Ioo 0 1), noon_realign_witness 1 q x y > 1 → 0 ≤ deriv (fun u => noon_realign_witness 1 q u y) x ∧ 0 ≤ deriv (fun v => noon_realign_witness 1 q x v) y) ∧ (∀ (q x y x' y' : ℝ) (hq : q ∈ Set.Ioo 0 1) (hx : 0 < x ∧ x ≤ x' ∧ x' < 1) (hy : 0 < y ∧ y ≤ y' ∧ y' < 1), noon_realign_witness 1 q x y > 1 → noon_realign_witness 1 q x' y' > 1) := by constructor
Under W_1 > 1, the quadratic G(q*x, (1-q)*y) is strictly positive, yielding linear lower bounds on the transmissivity combinations.
Setting s = q*x > 0 and r = (1-q)*y > 0, the witness deficit identity shows G(s, r) > 0, which implies 1 ≤ 2*s + 4*r and 1 ≤ 4*s + 2*r.
uses
mul_pos,op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one,op_1693_noon_witness_loss_monotonicity_one_G_pos_x_bound,op_1693_noon_witness_loss_monotonicity_one_G_pos_y_bound· intro q x y hq hx hy hW have hqx : 0 < q*x := mul_pos hq.1 hx.1 have hry : 0 < (1-q)*y := mul_pos (by linarith [hq.2]) hy.1 have hG : 0 < (q*x)^2 + ((1-q)*y)^2 + 4*(q*x)*((1-q)*y) - (q*x) - ((1-q)*y) := by have h := op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one q x y linarith [h, hW] have hineq : 1 ≤ 2*(q*x) + 4*((1-q)*y) := op_1693_noon_witness_loss_monotonicity_one_G_pos_x_bound (q*x) ((1-q)*y) hqx hry hG have hineq2 : 1 ≤ 4*(q*x) + 2*((1-q)*y) := op_1693_noon_witness_loss_monotonicity_one_G_pos_y_bound (q*x) ((1-q)*y) hqx hry hG
The partial derivative of the witness with respect to x is non-negative.
Substituting the derivative formula 2*q*(2*(q*x) + 4*((1-q)*y) - 1), non-negativity follows since q > 0 and 2*(q*x) + 4*((1-q)*y) - 1 ≥ 0.
uses
op_1693_noon_witness_loss_monotonicity_one_witness_one_x_derivconstructor · have hd := op_1693_noon_witness_loss_monotonicity_one_witness_one_x_deriv q y x rw [hd] nlinarith [hineq, hq.1]
The partial derivative of the witness with respect to y is non-negative.
Substituting the derivative formula 2*(1-q)*(4*(q*x) + 2*((1-q)*y) - 1), non-negativity follows since 1-q > 0 and 4*(q*x) + 2*((1-q)*y) - 1 ≥ 0.
uses
op_1693_noon_witness_loss_monotonicity_one_witness_one_y_deriv· have hd := op_1693_noon_witness_loss_monotonicity_one_witness_one_y_deriv q x y rw [hd] have hq1 : (0:ℝ) < 1 - q := by linarith [hq.2] nlinarith [hineq2, hq1]
For the second conjunct, assuming W_1(q, x, y) > 1, the quadratic G(q*x, (1-q)*y) is strictly positive.
We introduce the parameters and use the deficit identity to deduce G(q*x, (1-q)*y) > 0.
uses
mul_pos,op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one· intro q x y x' y' hq hx hy hW have hqx : 0 < q*x := mul_pos hq.1 hx.1 have hry : 0 < (1-q)*y := mul_pos (by linarith [hq.2]) hy.1 have hG : 0 < (q*x)^2 + ((1-q)*y)^2 + 4*(q*x)*((1-q)*y) - (q*x) - ((1-q)*y) := by have h := op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one q x y linarith [h, hW]
The witness condition is preserved at increased transmissivities: W_1(q, x', y') > 1.
Setting non-negative increments ds = q*(x'-x) ≥ 0 and dr = (1-q)*(y'-y) ≥ 0, monotonicity of G gives G(q*x', (1-q)*y') > 0, which by the deficit formula implies W_1(q, x', y') > 1.
uses
mul_nonneg,op_1693_noon_witness_loss_monotonicity_one_G_mono,op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_onehave hds : 0 ≤ q*(x'-x) := mul_nonneg hq.1.le (by linarith [hx.2.1]) have hdr : 0 ≤ (1-q)*(y'-y) := mul_nonneg (by linarith [hq.2]) (by linarith [hy.2.1]) have hmono := op_1693_noon_witness_loss_monotonicity_one_G_mono (q*x) ((1-q)*y) (q*(x'-x)) ((1-q)*(y'-y)) hqx hry hds hdr hG have hx' : q*x + q*(x'-x) = q*x' := by ring have hy' : (1-q)*y + (1-q)*(y'-y) = (1-q)*y' := by ring rw [hx', hy'] at hmono have hW' := op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one q x' y' linarith [hW', hmono]
Polynomial.deriv- The derivative of the polynomial evaluation function at x equals the evaluation of the formal polynomial derivative at x.
Finset.sum_range_succ- A sum over a range of length n + 1 equals the sum over range n plus the term at index n.
op_1693_noon_witness_loss_monotonicity_one_deriv_quadratic- The derivative of a quadratic function a u^2 + b u + c with respect to u at x is 2 a x + b.
sq_nonneg- The square of any real number is non-negative.
mul_nonneg- The product of two non-negative numbers is non-negative.
mul_pos- The product of two strictly positive numbers is strictly positive.
op_1693_noon_witness_loss_monotonicity_one_G_pos_x_bound- For positive s and r, if s^2 + r^2 + 4sr - s - r > 0, then 1 ≤ 2s + 4r.
op_1693_noon_witness_loss_monotonicity_one_G_pos_y_bound- For positive s and r, if s^2 + r^2 + 4sr - s - r > 0, then 1 ≤ 4s + 2r.
op_1693_noon_witness_loss_monotonicity_one_G_mono- If G(s, r) > 0 for positive s, r, then G(s + ds, r + dr) > 0 for any non-negative increments ds, dr.
op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one- The witness deficit W_1(q, x, y) - 1 equals 2((qx)^2 + ((1-q)y)^2 + 4(qx)((1-q)y) - qx - (1-q)y).
op_1693_noon_witness_loss_monotonicity_one_witness_one_x_deriv- The partial derivative of W_1 with respect to x is 2q(2(qx) + 4((1-q)y) - 1).
op_1693_noon_witness_loss_monotonicity_one_witness_one_y_deriv- The partial derivative of W_1 with respect to y is 2(1-q)(4(qx) + 2((1-q)y) - 1).
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
Choose a tactic — a marked line — to see the goals Lean had before and after it, as in Lean’s infoview. Step with ← → or j k.
private lemma op_1693_noon_witness_loss_monotonicity_one_deriv_quadratic (a b c x : ℝ) : deriv (fun u => a*u^2 + b*u + c) x = 2*a*x + b := by have h : (fun u => a*u^2 + b*u + c) = fun u => (Polynomial.C a * Polynomial.X^2 + Polynomial.C b * Polynomial.X + Polynomial.C c).eval u := by funext u; simp rw [h] simp only [Polynomial.deriv] simp ringprivate lemma op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one (q x y : ℝ) : noon_realign_witness 1 q x y - 1 = 2*((q*x)^2 + ((1-q)*y)^2 + 4*(q*x)*((1-q)*y) - (q*x) - ((1-q)*y)) := by unfold noon_realign_witness simp [Finset.sum_range_succ, Nat.choose] ringprivate lemma op_1693_noon_witness_loss_monotonicity_one_witness_one_x_deriv (q y x : ℝ) : deriv (fun u => noon_realign_witness 1 q u y) x = 2*q*(2*(q*x) + 4*((1-q)*y) - 1) := by have hfun : (fun u => noon_realign_witness 1 q u y) = fun u => (2*q^2)*u^2 + (-2*q + 8*q*(1-q)*y)*u + (1 - 2*((1-q)*y) + 2*((1-q)*y)^2) := by funext u unfold noon_realign_witness simp [Finset.sum_range_succ, Nat.choose] ring rw [hfun] rw [op_1693_noon_witness_loss_monotonicity_one_deriv_quadratic] ringprivate lemma op_1693_noon_witness_loss_monotonicity_one_witness_one_y_deriv (q x y : ℝ) : deriv (fun v => noon_realign_witness 1 q x v) y = 2*(1-q)*(4*(q*x) + 2*((1-q)*y) - 1) := by have hfun : (fun v => noon_realign_witness 1 q x v) = fun v => (2*(1-q)^2)*v^2 + (-2*(1-q) + 8*(1-q)*q*x)*v + (1 - 2*(q*x) + 2*(q*x)^2) := by funext v unfold noon_realign_witness simp [Finset.sum_range_succ, Nat.choose] ring rw [hfun] rw [op_1693_noon_witness_loss_monotonicity_one_deriv_quadratic] ringprivate lemma op_1693_noon_witness_loss_monotonicity_one_G_pos_x_bound (s r : ℝ) (hs : 0 < s) (hr : 0 < r) (hG : 0 < s^2 + r^2 + 4*s*r - s - r) : 1 ≤ 2*s + 4*r := by by_contra h push Not at h nlinarith [sq_nonneg s, sq_nonneg (r - 1), mul_nonneg hs.le (show (0:ℝ) ≤ 1 - 2*s - 4*r by linarith), mul_nonneg hr.le (show (0:ℝ) ≤ 1 - r by linarith)]private lemma op_1693_noon_witness_loss_monotonicity_one_G_pos_y_bound (s r : ℝ) (hs : 0 < s) (hr : 0 < r) (hG : 0 < s^2 + r^2 + 4*s*r - s - r) : 1 ≤ 4*s + 2*r := by by_contra h push Not at h nlinarith [sq_nonneg r, sq_nonneg (s - 1), mul_nonneg hr.le (show (0:ℝ) ≤ 1 - 4*s - 2*r by linarith), mul_nonneg hs.le (show (0:ℝ) ≤ 1 - s by linarith)]private lemma op_1693_noon_witness_loss_monotonicity_one_G_mono (s r ds dr : ℝ) (hs : 0 < s) (hr : 0 < r) (hds : 0 ≤ ds) (hdr : 0 ≤ dr) (hG : 0 < s^2 + r^2 + 4*s*r - s - r) : 0 < (s+ds)^2 + (r+dr)^2 + 4*(s+ds)*(r+dr) - (s+ds) - (r+dr) := by have hA : 0 ≤ 2*s + 4*r - 1 := by have := op_1693_noon_witness_loss_monotonicity_one_G_pos_x_bound s r hs hr hG; linarith have hB : 0 ≤ 4*s + 2*r - 1 := by have := op_1693_noon_witness_loss_monotonicity_one_G_pos_y_bound s r hs hr hG; linarith nlinarith [mul_nonneg hA hds, mul_nonneg hB hdr, mul_nonneg hds hdr, sq_nonneg ds, sq_nonneg dr]/-- For photon number N = 1, whenever W_1 > 1 the partial derivatives are non-negative and the detected region is upward closed. -/theorem op_1693_noon_witness_loss_monotonicity_one : (∀ (q x y : ℝ) (hq : q ∈ Set.Ioo 0 1) (hx : x ∈ Set.Ioo 0 1) (hy : y ∈ Set.Ioo 0 1), noon_realign_witness 1 q x y > 1 → 0 ≤ deriv (fun u => noon_realign_witness 1 q u y) x ∧ 0 ≤ deriv (fun v => noon_realign_witness 1 q x v) y) ∧ (∀ (q x y x' y' : ℝ) (hq : q ∈ Set.Ioo 0 1) (hx : 0 < x ∧ x ≤ x' ∧ x' < 1) (hy : 0 < y ∧ y ≤ y' ∧ y' < 1), noon_realign_witness 1 q x y > 1 → noon_realign_witness 1 q x' y' > 1) := by constructor · intro q x y hq hx hy hW have hqx : 0 < q*x := mul_pos hq.1 hx.1 have hry : 0 < (1-q)*y := mul_pos (by linarith [hq.2]) hy.1 have hG : 0 < (q*x)^2 + ((1-q)*y)^2 + 4*(q*x)*((1-q)*y) - (q*x) - ((1-q)*y) := by have h := op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one q x y linarith [h, hW] have hineq : 1 ≤ 2*(q*x) + 4*((1-q)*y) := op_1693_noon_witness_loss_monotonicity_one_G_pos_x_bound (q*x) ((1-q)*y) hqx hry hG have hineq2 : 1 ≤ 4*(q*x) + 2*((1-q)*y) := op_1693_noon_witness_loss_monotonicity_one_G_pos_y_bound (q*x) ((1-q)*y) hqx hry hG constructor · have hd := op_1693_noon_witness_loss_monotonicity_one_witness_one_x_deriv q y x rw [hd] nlinarith [hineq, hq.1] · have hd := op_1693_noon_witness_loss_monotonicity_one_witness_one_y_deriv q x y rw [hd] have hq1 : (0:ℝ) < 1 - q := by linarith [hq.2] nlinarith [hineq2, hq1] · intro q x y x' y' hq hx hy hW have hqx : 0 < q*x := mul_pos hq.1 hx.1 have hry : 0 < (1-q)*y := mul_pos (by linarith [hq.2]) hy.1 have hG : 0 < (q*x)^2 + ((1-q)*y)^2 + 4*(q*x)*((1-q)*y) - (q*x) - ((1-q)*y) := by have h := op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one q x y linarith [h, hW] have hds : 0 ≤ q*(x'-x) := mul_nonneg hq.1.le (by linarith [hx.2.1]) have hdr : 0 ≤ (1-q)*(y'-y) := mul_nonneg (by linarith [hq.2]) (by linarith [hy.2.1]) have hmono := op_1693_noon_witness_loss_monotonicity_one_G_mono (q*x) ((1-q)*y) (q*(x'-x)) ((1-q)*(y'-y)) hqx hry hds hdr hG have hx' : q*x + q*(x'-x) = q*x' := by ring have hy' : (1-q)*y + (1-q)*(y'-y) = (1-q)*y' := by ring rw [hx', hy'] at hmono have hW' := op_1693_noon_witness_loss_monotonicity_one_witness_one_minus_one q x' y' linarith [hW', hmono]