AFTD Auto-Formalizing Theoretical Domains

plank_system_sign_vector_exists

theoremverifiedformalization

Let $V$ be a real inner product space, let $\iota$ be a finite type, let $n : \iota \to V$ be unit vectors, let $w : \iota \to \mathbb{R}$ be positive widths ($w_i > 0$ for all $i$), and let $s : \iota \to \mathbb{R}$ be shifts. Then there exists $\varepsilon : \iota \to \mathbb{R}$ such that for every $i \in \iota$, $\varepsilon_i = 1$ or $\varepsilon_i = -1$, and the vector $z = \frac{1}{2} \sum_{i \in \iota} (\varepsilon_i w_i) \cdot n_i$ satisfies $w_i / 2 \le |\langle n_i, z \rangle - s_i|$ for all $i \in \iota$.

Statement

open scoped RealInnerProductSpace in
theorem plank_system_sign_vector_exists
    {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V]
    {ι : Type*} [Fintype ι]
    (n : ι → V) (hn : ∀ i, ‖n i‖ = 1)
    (w : ι → ℝ) (hw : ∀ i, 0 < w i)
    (s : ι → ℝ) :
    ∃ ε : ι → ℝ, (∀ i, ε i = 1 ∨ ε i = -1) ∧
      let z := (1 / 2 : ℝ) • ∑ i, (ε i * w i) • n i
      ∀ i, w i / 2 ≤ |⟪n i, z⟫ - s i|
Formalizes
arXiv:2611.00379, Lemma 3 (A sharp plank theorem for cube annuli)
Verified
04 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 1
  • plank_energy_diff_single_flipLet $V$ be a real inner product space and $\iota$ a finite type with decidable equality. Let $n : \iota \to V$ be unit vectors,…

Read back from the Lean

Let $V$ be a real inner product space (equipped with [NormedAddCommGroup V] and [InnerProductSpace ℝ V]), and let $\iota$ be a finite type ([Fintype ι]). Given: 1. a family of unit vectors $n : \iota \to V$ satisfying $\|n(i)\| = 1$ for all $i \in \iota$, 2. a family of positive weights/widths $w : \iota \to \mathbb{R}$ satisfying $w(i) > 0$ for all $i \in \iota$, and 3. a family of shifts $s : \iota \to \mathbb{R}$, there exists a sign function $\varepsilon : \iota \to \mathbb{R}$ such that: - for every $i \in \iota$, $\varepsilon(i) = 1$ or $\varepsilon(i) = -1$, and - defining $z \in V$ by $z = \frac{1}{2} \sum_{i \in \iota} (\varepsilon(i) w(i)) \cdot n(i)$, for every $i \in \iota$, $$\frac{w(i)}{2} \le |\langle n(i), z \rangle - s(i)|$$ (where $\langle \cdot, \cdot \rangle$ is the real inner product on $V$).

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 existence of the sign vector is proved using a discrete variational argument (the energy method for Bang's plank problem). An energy functional E(ε) = ‖z(ε)‖² - ∑_j ε_j w_j s_j is defined on the finite set of sign assignments ε ∈ {±1}^ι, which achieves its maximum at some configuration. Perturbing this maximizer by flipping a single coordinate ε_i to -ε_i cannot increase the energy, and the single-flip energy difference formula implies that this non-increase is equivalent to ε_i (⟨n_i, z⟩ - s_i) ≥ w_i / 2; bounding the left-hand side by |⟨n_i, z⟩ - s_i| yields the desired plank inequality.

  1. Define an energy functional on sign configurations and choose a configuration b that maximizes it.

    For each sign assignment sgn : ι → ℝ, define the center vector z(sgn) = (1/2) ∑_j (sgn_j * w_j) • n_j and the energy E(sgn, pt) = ‖pt‖² - ∑_j sgn_j * w_j * s_j. By identifying sign vectors with boolean functions b : ι → Bool (setting ε_j = 1 if b(j) = true and -1 otherwise), the set of all configurations is finite and nonempty. By `Finset.exists_max_image`, there exists a configuration b that maximizes the cost E(ε, z(ε)).

    lines 12–21

    uses Finset.mem_univ, Finset.exists_max_image

      classical
      let z_of := fun (sgn : ι → ℝ) => (1 / 2 : ℝ) • ∑ j, (sgn j * w j) • n j
      let E := fun (sgn : ι → ℝ) (pt : V) => ‖pt‖ ^ 2 - ∑ j, sgn j * w j * s j
      let cost := fun (b : ι → Bool) =>
        let ε := fun j => if b j then (1 : ℝ) else -1
        E ε (z_of ε)
      have h_nonempty : (Finset.univ : Finset (ι → Bool)).Nonempty :=
        ⟨fun _ => true, Finset.mem_univ _⟩
      obtain ⟨b, -, hb⟩ := Finset.exists_max_image Finset.univ cost h_nonempty
      let ε := fun j => if b j then (1 : ℝ) else -1
  2. Provide ε as the witness sign vector and show that ε_i ∈ {1, -1} for each i.

    Define ε to be the sign vector associated with the maximizer b. For each index i, casing on whether b(i) is true or false confirms that ε_i = 1 or ε_i = -1.

    lines 22–25

      refine ⟨ε, ?_, ?_⟩
      · intro i
        dsimp [ε]
        cases b i <;> simp
  3. Flipping the i-th sign gives an alternate configuration b' whose cost does not exceed that of b.

    Fix an index i, and let b' be the configuration obtained by flipping the boolean value at i. Since b maximizes cost over all configurations, cost(b') ≤ cost(b), so cost(b) - cost(b') ≥ 0 by `sub_nonneg.mpr`. Moreover, the signs corresponding to b' agree with ε at all j ≠ i and equal -ε_i at i.

    lines 26–38

    uses Finset.mem_univ, sub_nonneg.mpr

      · intro z i
        let b' := Function.update b i (!b i)
        have hle : cost b' ≤ cost b := hb b' (Finset.mem_univ _)
        have h_nonneg : 0 ≤ cost b - cost b' := sub_nonneg.mpr hle
        have hε : ε i = 1 ∨ ε i = -1 := by
          dsimp [ε]
          cases b i <;> simp
        have h_upd : (fun j => if b' j then (1 : ℝ) else -1) = Function.update ε i (-ε i) := by
          ext j
          by_cases hj : j = i
          · simp [b', ε, hj]
            cases b i <;> simp
          · simp [b', ε, hj]
  4. The single-flip energy formula implies ε_i * (⟨n_i, z⟩ - s_i) - w_i / 2 ≥ 0.

    By the lemma `plank_energy_diff_single_flip`, flipping the sign ε_i to -ε_i changes the energy by exactly 2 * w_i * (ε_i * (⟨n_i, z_of(ε)⟩ - s_i) - w_i / 2). Since cost(b) - cost(b') ≥ 0 and 2 * w_i > 0 (as w_i > 0), dividing by 2 * w_i yields ε_i * (⟨n_i, z_of(ε)⟩ - s_i) - w_i / 2 ≥ 0.

    lines 39–47

    uses plank_energy_diff_single_flip

        have hdiff := plank_energy_diff_single_flip n hn w s ε i hε
        have heq : cost b - cost b' = 2 * w i * (ε i * (⟪n i, z_of ε⟫ - s i) - w i / 2) := by
          dsimp [cost]
          rw [h_upd]
          exact hdiff
        have h_pos : 0 < 2 * w i := by linarith [hw i]
        have h_ge : 0 ≤ ε i * (⟪n i, z_of ε⟫ - s i) - w i / 2 := by
          rw [heq] at h_nonneg
          nlinarith
  5. The plank inequality w_i / 2 ≤ |⟨n_i, z⟩ - s_i| holds for each i.

    Since ε_i ∈ {1, -1}, if ε_i = 1 then ε_i * (⟨n_i, z⟩ - s_i) = ⟨n_i, z⟩ - s_i ≤ |⟨n_i, z⟩ - s_i| using `le_abs_self`; if ε_i = -1 then ε_i * (⟨n_i, z⟩ - s_i) = -(⟨n_i, z⟩ - s_i) ≤ |⟨n_i, z⟩ - s_i| using `neg_le_abs`. Combining this upper bound with the previous inequality w_i / 2 ≤ ε_i * (⟨n_i, z⟩ - s_i) establishes w_i / 2 ≤ |⟨n_i, z⟩ - s_i|.

    lines 48–55

    uses le_abs_self, neg_mul, one_mul, neg_le_abs

        have h_le_abs : ε i * (⟪n i, z_of ε⟫ - s i) ≤ |⟪n i, z_of ε⟫ - s i| := by
          rcases hε with h1 | h2
          · simp [h1, le_abs_self]
          · rw [h2, neg_mul, one_mul]
            exact neg_le_abs (⟪n i, z_of ε⟫ - s i)
        have hz : z = z_of ε := rfl
        rw [hz]
        linarith
Finset.mem_univ
Every element of a finite type belongs to the universal finset `Finset.univ`.
Finset.exists_max_image
A real-valued (or linearly ordered) function on a nonempty finite set attains its maximum value at some element.
sub_nonneg.mpr
If b ≤ a, then 0 ≤ a - b.
plank_energy_diff_single_flip
Flipping the sign ε_i to -ε_i changes the plank energy E(ε, z) by 2 * w_i * (ε_i * (⟨n_i, z⟩ - s_i) - w_i / 2).
le_abs_self
Every element is less than or equal to its absolute value: a ≤ |a|.
neg_mul
Negation distributes over multiplication on the left: (-a) * b = -(a * b).
one_mul
Multiplying by 1 on the left is the identity: 1 * a = a.
neg_le_abs
The negation of any element is less than or equal to its absolute value: -a ≤ |a|.

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 RealInnerProductSpace in/-- Sign vector existence lemma for plank systems. -/theorem plank_system_sign_vector_exists    {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V]    {ι : Type*} [Fintype ι]    (n : ι → V) (hn : ∀ i, ‖n i‖ = 1)    (w : ι → ℝ) (hw : ∀ i, 0 < w i)    (s : ι → ℝ) :    ∃ ε : ι → ℝ, (∀ i, ε i = 1 ∨ ε i = -1) ∧      let z := (1 / 2 : ℝ) • ∑ i, (ε i * w i) • n i      ∀ i, w i / 2 ≤ |⟪n i, z⟫ - s i| := by  classical  let z_of := fun (sgn : ι → ℝ) => (1 / 2 : ℝ) • ∑ j, (sgn j * w j) • n j  let E := fun (sgn : ι → ℝ) (pt : V) => ‖pt‖ ^ 2 - ∑ j, sgn j * w j * s j  let cost := fun (b : ι → Bool) =>    let ε := fun j => if b j then (1 : ℝ) else -1    E ε (z_of ε)  have h_nonempty : (Finset.univ : Finset (ι → Bool)).Nonempty :=    ⟨fun _ => true, Finset.mem_univ _⟩  obtain ⟨b, -, hb⟩ := Finset.exists_max_image Finset.univ cost h_nonempty  let ε := fun j => if b j then (1 : ℝ) else -1  refine ⟨ε, ?_, ?_⟩  · intro i    dsimp [ε]    cases b i <;> simp  · intro z i    let b' := Function.update b i (!b i)    have hle : cost b' ≤ cost b := hb b' (Finset.mem_univ _)    have h_nonneg : 0 ≤ cost b - cost b' := sub_nonneg.mpr hle    have hε : ε i = 1 ∨ ε i = -1 := by      dsimp [ε]      cases b i <;> simp    have h_upd : (fun j => if b' j then (1 : ℝ) else -1) = Function.update ε i (-ε i) := by      ext j      by_cases hj : j = i      · simp [b', ε, hj]        cases b i <;> simp      · simp [b', ε, hj]    have hdiff := plank_energy_diff_single_flip n hn w s ε i hε    have heq : cost b - cost b' = 2 * w i * (ε i * (⟪n i, z_of ε⟫ - s i) - w i / 2) := by      dsimp [cost]      rw [h_upd]      exact hdiff    have h_pos : 0 < 2 * w i := by linarith [hw i]    have h_ge : 0 ≤ ε i * (⟪n i, z_of ε⟫ - s i) - w i / 2 := by      rw [heq] at h_nonneg      nlinarith    have h_le_abs : ε i * (⟪n i, z_of ε⟫ - s i) ≤ |⟪n i, z_of ε⟫ - s i| := by      rcases hε with h1 | h2      · simp [h1, le_abs_self]      · rw [h2, neg_mul, one_mul]        exact neg_le_abs (⟪n i, z_of ε⟫ - s i)    have hz : z = z_of ε := rfl    rw [hz]    linarith
See it in the graphDiscuss this resultChallenge it