AFTD Auto-Formalizing Theoretical Domains

op_1172_optimal_constant_potential_function_exists

theoremverifiedformalization

There exist an activation budget f ∈ (0, 1), descent parameters η > 0 and G₀ ≥ 1, a constant σ_star ∈ (1, 2) satisfying σ_star³ - 19 * σ_star² + 27 * σ_star + 1 = 0, a target satisfaction threshold σ > σ_star, a margin c > 0, memory bounds 0 < H_min < f, a satisfaction lower bound 0 < varsigma_min < σ, and a potential function Φ : ℝ × ℝ → ℝ of class C¹ on the open positive quadrant (0, ∞) × (0, ∞) such that for all points p in the closed rectangle [H_min, f] × [varsigma_min, σ], the Fréchet derivative of Φ at p evaluates to at least c on the descent drift field descent_drift_field η G₀ p and to at least c on the repair drift field repair_drift_field f p.

Statement

theorem op_1172_optimal_constant_potential_function_exists :
  ∃ (η G₀ f σ c : ℝ) (H_min varsigma_min : ℝ),
    0 < η ∧ 1 ≤ G₀ ∧ 0 < f ∧ f < 1 ∧
    0 < H_min ∧ H_min < f ∧ 0 < varsigma_min ∧ varsigma_min < σ ∧
    (∃ σ_star : ℝ, 1 < σ_star ∧ σ_star < 2 ∧ σ_star^3 - 19 * σ_star^2 + 27 * σ_star + 1 = 0 ∧ σ_star < σ) ∧
    0 < c ∧
    ∃ (Φ : ℝ × ℝ → ℝ),
      ContDiffOn ℝ 1 Φ (Set.Ioi (0 : ℝ) ×ˢ Set.Ioi (0 : ℝ)) ∧
      ∀ p ∈ Set.Icc H_min f ×ˢ Set.Icc varsigma_min σ,
        c ≤ (fderiv ℝ Φ p) (descent_drift_field η G₀ p) ∧
        c ≤ (fderiv ℝ Φ p) (repair_drift_field f p)
Formalizes
arXiv:2611.06076 after Theorem 5.8
Verified
08 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 2
  • descent_drift_fieldThe continuous-time descent drift vector field at memory weight H and scaled satisfaction ς with descent rate η and threshold G₀ is (-η *…
  • repair_drift_fieldThe continuous-time repair drift vector field at memory weight H and scaled satisfaction ς with activation budget f is (f - H, 1 + H - ς).

Read back from the Lean

There exist real numbers $\eta$, $G_0$, $f$, $\sigma$, $c$, $H_{\min}$, and $\varsigma_{\min}$ such that: 1. $\eta > 0$, 2. $G_0 \ge 1$, 3. $0 < f < 1$, 4. $0 < H_{\min} < f$, 5. $0 < \varsigma_{\min} < \sigma$, 6. there exists a real number $\sigma^*$ such that $1 < \sigma^* < 2$, $(\sigma^*)^3 - 19(\sigma^*)^2 + 27\sigma^* + 1 = 0$, and $\sigma^* < \sigma$, 7. $c > 0$, and 8. there exists a function $\Phi : \mathbb{R} \times \mathbb{R} \to \mathbb{R}$ that is continuously differentiable of class $C^1$ on $(0, \infty) \times (0, \infty)$, such that for all $p = (p_1, p_2) \in [H_{\min}, f] \times [\varsigma_{\min}, \sigma]$: - the Fréchet derivative of $\Phi$ at $p$ evaluated on the vector $(-\eta p_1, \eta(p_2 - G_0))$ is at least $c$, and - the Fréchet derivative of $\Phi$ at $p$ evaluated on the vector $(f - p_1, 1 + p_1 - p_2)$ is at least $c$.

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

The theorem is proved by providing explicit numerical constants and a linear Lyapunov potential function. We choose the descent parameter η = 1, drift target G₀ = 3, activation budget f = 1/2, satisfaction threshold σ = 3, margin c = 1/4, memory lower bound H_min = 1/4, and satisfaction lower bound varsigma_min = 2. Applying the intermediate value theorem to the polynomial P(x) = x³ - 19x² + 27x + 1 on [1, 2], where P(1) = 10 > 0 and P(2) = -13 < 0, yields an intermediate root σ_star ∈ (1, 2) with σ_star < σ. Finally, the linear potential Φ(x, y) = -x - y is continuously differentiable everywhere, and linear arithmetic verifies that its Fréchet derivative evaluates to at least 1/4 on both the descent drift field and the repair drift field across the compact domain [1/4, 1/2] × [2, 3].

  1. The polynomial P(x) = x³ - 19x² + 27x + 1 is continuous on ℝ. automation

    Unfolding the definition of P, continuity follows because polynomial expressions formed from addition, subtraction, multiplication, and powers are continuous on ℝ.

    lines 9–10

      unfold op_1172_optimal_constant_potential_function_exists_poly
      continuity
  2. The parameter choices η = 1, G₀ = 3, f = 1/2, σ = 3, c = 1/4, H_min = 1/4, and varsigma_min = 2 satisfy the basic numerical inequalities required by the theorem. automation

    We supply the concrete witnesses η = 1, G₀ = 3, f = 1/2, σ = 3, c = 1/4, H_min = 1/4, and varsigma_min = 2. Automated evaluation (norm_num) verifies the nine scalar inequalities: 0 < 1, 1 ≤ 3, 0 < 1/2, 1/2 < 1, 0 < 1/4, 1/4 < 1/2, 0 < 2, 2 < 3, and 0 < 1/4.

    lines 24–26

      refine ⟨1, 3, 1/2, 3, 1/4, 1/4, 2, ?_⟩
      refine ⟨by norm_num, by norm_num, by norm_num, by norm_num,
              by norm_num, by norm_num, by norm_num, by norm_num, ?_, by norm_num, ?_⟩
  3. The intermediate value theorem applies to P on [1, 2], so the image P((1, 2)) contains the open interval (P(2), P(1)).

    Since 1 ≤ 2 and P is continuous on [1, 2] by restriction of its global continuity, the intermediate value theorem intermediate_value_Ioo' establishes that (P(2), P(1)) ⊆ P((1, 2)).

    lines 27–30

    uses intermediate_value_Ioo'

      · have h12 : (1 : ℝ) ≤ 2 := by norm_num
        have hcont : ContinuousOn op_1172_optimal_constant_potential_function_exists_poly (Set.Icc 1 2) :=
          op_1172_optimal_constant_potential_function_exists_poly_cont.continuousOn
        have h_ivt := intermediate_value_Ioo' h12 hcont
  4. The value 0 belongs to the interval (P(2), P(1)). automation

    Evaluating the polynomial at the endpoints gives P(2) = 2³ - 19·2² + 27·2 + 1 = -13 < 0 and P(1) = 1³ - 19·1² + 27·1 + 1 = 10 > 0. By Set.mem_Ioo, this confirms 0 ∈ (-13, 10) = (P(2), P(1)).

    lines 31–37

    uses Set.mem_Ioo

        have h0 : (0 : ℝ) ∈ Set.Ioo (op_1172_optimal_constant_potential_function_exists_poly 2)
                                   (op_1172_optimal_constant_potential_function_exists_poly 1) := by
          rw [Set.mem_Ioo]
          change (op_1172_optimal_constant_potential_function_exists_poly 2 < 0 ∧
                  0 < op_1172_optimal_constant_potential_function_exists_poly 1)
          unfold op_1172_optimal_constant_potential_function_exists_poly
          constructor <;> norm_num
  5. There exists σ_star ∈ (1, 2) such that P(σ_star) = 0 and σ_star < σ.

    Applying the intermediate value theorem to 0 ∈ (P(2), P(1)) via Set.mem_image yields a root σ_star ∈ (1, 2) satisfying P(σ_star) = 0. Since σ_star < 2, linear arithmetic gives σ_star < 3 = σ.

    lines 38–42

    uses Set.mem_image, Set.mem_Ioo

        have h_ex := h_ivt h0
        rw [Set.mem_image] at h_ex
        rcases h_ex with ⟨σ_star, hσ_mem, hσ_val⟩
        rw [Set.mem_Ioo] at hσ_mem
        refine ⟨σ_star, hσ_mem.1, hσ_mem.2, hσ_val, by linarith [hσ_mem.2]⟩
  6. The potential function Φ(x, y) = -x - y is of class C¹ on the open positive quadrant (0, ∞) × (0, ∞).

    We instantiate the potential function Φ with the continuous linear map (x, y) ↦ -x - y. As a continuous linear map on ℝ × ℝ, Φ is smooth everywhere, and restricting to the positive quadrant preserves C¹ regularity.

    lines 43–44

      · refine ⟨op_1172_optimal_constant_potential_function_exists_phi,
                op_1172_optimal_constant_potential_function_exists_phi.contDiff.contDiffOn, ?_⟩
  7. For any p = (x, y) ∈ [1/4, 1/2] × [2, 3], we have 1/4 ≤ x ≤ 1/2 and 2 ≤ y ≤ 3, and it suffices to bound the derivative of Φ on the descent and repair drift fields separately.

    We introduce p ∈ [1/4, 1/2] × [2, 3] and unpack the rectangle membership Set.mem_Icc into the coordinate bounds 1/4 ≤ p.1 ≤ 1/2 and 2 ≤ p.2 ≤ 3, splitting the goal into certifying the descent drift field and the repair drift field.

    lines 45–48

    uses Set.mem_Icc

        intro p hp
        rcases hp with ⟨h1, h2⟩
        rw [Set.mem_Icc] at h1 h2
        constructor
  8. The Fréchet derivative of Φ at p evaluated on the descent drift field is at least 1/4. automation

    By op_1172_optimal_constant_potential_function_exists_phi.fderiv, the Fréchet derivative of the linear map Φ at p is Φ itself. Evaluating Φ on the descent drift field (-p.1, p.2 - 3) yields -(-p.1) - (p.2 - 3) = p.1 - p.2 + 3. Given p.1 ≥ 1/4 and p.2 ≤ 3, linear arithmetic (linarith) proves p.1 - p.2 + 3 ≥ 1/4 - 3 + 3 = 1/4.

    lines 49–52

    uses op_1172_optimal_constant_potential_function_exists_phi.fderiv

        · rw [op_1172_optimal_constant_potential_function_exists_phi.fderiv]
          change 1/4 ≤ - (descent_drift_field 1 3 p).1 - (descent_drift_field 1 3 p).2
          dsimp [descent_drift_field]
          linarith
  9. The Fréchet derivative of Φ at p evaluated on the repair drift field is at least 1/4. automation

    Using op_1172_optimal_constant_potential_function_exists_phi.fderiv again, the Fréchet derivative of Φ at p is Φ itself. Evaluating Φ on the repair drift field (1/2 - p.1, 1 + p.1 - p.2) gives -(1/2 - p.1) - (1 + p.1 - p.2) = p.2 - 3/2. Since p.2 ≥ 2, linear arithmetic (linarith) proves p.2 - 3/2 ≥ 2 - 3/2 = 1/2 ≥ 1/4.

    lines 53–56

    uses op_1172_optimal_constant_potential_function_exists_phi.fderiv

        · rw [op_1172_optimal_constant_potential_function_exists_phi.fderiv]
          change 1/4 ≤ - (repair_drift_field (1/2) p).1 - (repair_drift_field (1/2) p).2
          dsimp [repair_drift_field]
          linarith
intermediate_value_Ioo'
If f is continuous on [a, b] with a ≤ b, then the open interval (f(b), f(a)) is contained in the image f((a, b)).
Set.mem_Ioo
An element x belongs to the open interval (a, b) if and only if a < x < b.
Set.mem_image
An element y belongs to the image f(s) if and only if there exists x ∈ s such that f(x) = y.
Set.mem_Icc
An element x belongs to the closed interval [a, b] if and only if a ≤ x ≤ b.
op_1172_optimal_constant_potential_function_exists_phi.fderiv
The Fréchet derivative of the continuous linear map Φ at any point x equals Φ itself.

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

def op_1172_optimal_constant_potential_function_exists_phi : (ℝ × ℝ) →L[ℝ] ℝ :=  (-ContinuousLinearMap.fst ℝ ℝ ℝ) - ContinuousLinearMap.snd ℝ ℝ ℝdef op_1172_optimal_constant_potential_function_exists_poly (x : ℝ) : ℝ :=  x^3 - 19 * x^2 + 27 * x + 1lemma op_1172_optimal_constant_potential_function_exists_poly_cont :    Continuous op_1172_optimal_constant_potential_function_exists_poly := by  unfold op_1172_optimal_constant_potential_function_exists_poly  continuity/-- Existence of a continuously differentiable Lyapunov potential function certifying drift above sigma_star. -/theorem op_1172_optimal_constant_potential_function_exists :  ∃ (η G₀ f σ c : ℝ) (H_min varsigma_min : ℝ),    0 < η ∧ 1 ≤ G₀ ∧ 0 < f ∧ f < 1 ∧    0 < H_min ∧ H_min < f ∧ 0 < varsigma_min ∧ varsigma_min < σ ∧    (∃ σ_star : ℝ, 1 < σ_star ∧ σ_star < 2 ∧ σ_star^3 - 19 * σ_star^2 + 27 * σ_star + 1 = 0 ∧ σ_star < σ) ∧    0 < c ∧    ∃ (Φ : ℝ × ℝ → ℝ),      ContDiffOn ℝ 1 Φ (Set.Ioi (0 : ℝ) ×ˢ Set.Ioi (0 : ℝ)) ∧      ∀ p ∈ Set.Icc H_min f ×ˢ Set.Icc varsigma_min σ,        c ≤ (fderiv ℝ Φ p) (descent_drift_field η G₀ p) ∧        c ≤ (fderiv ℝ Φ p) (repair_drift_field f p) := by  refine ⟨1, 3, 1/2, 3, 1/4, 1/4, 2, ?_⟩  refine ⟨by norm_num, by norm_num, by norm_num, by norm_num,          by norm_num, by norm_num, by norm_num, by norm_num, ?_, by norm_num, ?_⟩  · have h12 : (1 : ℝ) ≤ 2 := by norm_num    have hcont : ContinuousOn op_1172_optimal_constant_potential_function_exists_poly (Set.Icc 1 2) :=      op_1172_optimal_constant_potential_function_exists_poly_cont.continuousOn    have h_ivt := intermediate_value_Ioo' h12 hcont    have h0 : (0 : ℝ) ∈ Set.Ioo (op_1172_optimal_constant_potential_function_exists_poly 2)                               (op_1172_optimal_constant_potential_function_exists_poly 1) := by      rw [Set.mem_Ioo]      change (op_1172_optimal_constant_potential_function_exists_poly 2 < 0 ∧              0 < op_1172_optimal_constant_potential_function_exists_poly 1)      unfold op_1172_optimal_constant_potential_function_exists_poly      constructor <;> norm_num    have h_ex := h_ivt h0    rw [Set.mem_image] at h_ex    rcases h_ex with ⟨σ_star, hσ_mem, hσ_val⟩    rw [Set.mem_Ioo] at hσ_mem    refine ⟨σ_star, hσ_mem.1, hσ_mem.2, hσ_val, by linarith [hσ_mem.2]⟩  · refine ⟨op_1172_optimal_constant_potential_function_exists_phi,            op_1172_optimal_constant_potential_function_exists_phi.contDiff.contDiffOn, ?_⟩    intro p hp    rcases hp with ⟨h1, h2⟩    rw [Set.mem_Icc] at h1 h2    constructor    · rw [op_1172_optimal_constant_potential_function_exists_phi.fderiv]      change 1/4 ≤ - (descent_drift_field 1 3 p).1 - (descent_drift_field 1 3 p).2      dsimp [descent_drift_field]      linarith    · rw [op_1172_optimal_constant_potential_function_exists_phi.fderiv]      change 1/4 ≤ - (repair_drift_field (1/2) p).1 - (repair_drift_field (1/2) p).2      dsimp [repair_drift_field]      linarith
See it in the graphDiscuss this resultChallenge it