AFTD Auto-Formalizing Theoretical Domains

plank_energy_diff_single_flip

theoremverified

Let $V$ be a real inner product space and $\iota$ a finite type with decidable equality. Let $n : \iota \to V$ be unit vectors, $w : \iota \to \mathbb{R}$ widths, $s : \iota \to \mathbb{R}$ shifts, $\varepsilon : \iota \to \mathbb{R}$ signs, and $i \in \iota$ with $\varepsilon_i = 1$ or $\varepsilon_i = -1$. Let $\varepsilon'$ be $\varepsilon$ with the sign at $i$ flipped, let $z = \frac{1}{2}\sum_j (\varepsilon_j w_j) \cdot n_j$ and $z' = \frac{1}{2}\sum_j (\varepsilon'_j w_j) \cdot n_j$. Define the energy $E(\sigma, v) = \|v\|^2 - \sum_j \sigma_j w_j s_j$. Then $E(\varepsilon, z) - E(\varepsilon', z') = 2 w_i (\varepsilon_i (\langle n_i, z \rangle - s_i) - w_i / 2)$.

Statement

open scoped RealInnerProductSpace in
lemma plank_energy_diff_single_flip
    {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V]
    {ι : Type*} [Fintype ι] [DecidableEq ι]
    (n : ι → V) (hn : ∀ i, ‖n i‖ = 1)
    (w : ι → ℝ) (s : ι → ℝ) (ε : ι → ℝ) (i : ι) (hε : ε i = 1 ∨ ε i = -1) :
    let ε' := Function.update ε i (-ε i)
    let z := (1 / 2 : ℝ) • ∑ j, (ε j * w j) • n j
    let z' := (1 / 2 : ℝ) • ∑ j, (ε' j * w j) • n j
    let E := fun (sgn : ι → ℝ) (pt : V) => ‖pt‖ ^ 2 - ∑ j, sgn j * w j * s j
    E ε z - E ε' z' = 2 * w i * (ε i * (⟪n i, z⟫ - s i) - w i / 2)
Source
arXiv:2611.00379, Lemma 3 proof
Verified
04 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 2
  • norm_sub_smul_sqLet $V$ be a real inner product space, let $z \in V$, let $n \in V$ be a unit vector with $\|n\| = 1$, and let $a \in \mathbb{R}$. Then…
  • plank_signed_sum_flipLet $V$ be a real vector space and $\iota$ a finite type with decidable equality. Given vectors $n : \iota \to V$, weights…
Used by 1
  • plank_system_sign_vector_existsLet $V$ be a real inner product space, let $\iota$ be a finite type, let $n : \iota \to V$ be unit vectors, let $w : \iota \to \mathbb{R}$…

Read back from the Lean

Let $V$ be a real inner product space, $\iota$ a finite type with decidable equality, $n : \iota \to V$ a family of vectors satisfying $\|n(i)\| = 1$ for all $i \in \iota$, and let $w, s, \varepsilon : \iota \to \mathbb{R}$. Let $i \in \iota$ be an index such that $\varepsilon(i) = 1$ or $\varepsilon(i) = -1$. Define: - $\varepsilon' : \iota \to \mathbb{R}$ as the function update of $\varepsilon$ at index $i$ to $-\varepsilon(i)$; - $z = \frac{1}{2} \sum_{j \in \iota} (\varepsilon(j) w(j)) n(j) \in V$; - $z' = \frac{1}{2} \sum_{j \in \iota} (\varepsilon'(j) w(j)) n(j) \in V$; - $E(\mathrm{sgn}, \mathrm{pt}) = \|\mathrm{pt}\|^2 - \sum_{j \in \iota} \mathrm{sgn}(j) w(j) s(j)$ for $\mathrm{sgn} : \iota \to \mathbb{R}$ and $\mathrm{pt} \in V$. Then: $$E(\varepsilon, z) - E(\varepsilon', z') = 2 w(i) \left(\varepsilon(i) (\langle n(i), z \rangle - s(i)) - \frac{w(i)}{2}\right).$$

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 proof computes the energy difference $E(\varepsilon, z) - E(\varepsilon', z')$ by analyzing how flipping a single sign $\varepsilon_i \in \{-1, 1\}$ affects each term in the definition of the energy. First, flipping the sign at $i$ changes the linear shift sum by $-2 \varepsilon_i w_i s_i$, as all other terms remain identical. Second, the center vector shifts by $z' = z - (\varepsilon_i w_i) n_i$, so expanding $\|z'\|^2$ with the unit vector property $\|n_i\| = 1$ and $(\varepsilon_i w_i)^2 = w_i^2$ yields $\|z\|^2 - 2 \varepsilon_i w_i \langle n_i, z \rangle + w_i^2$. Subtracting $E(\varepsilon', z')$ from $E(\varepsilon, z)$ cancels both $\|z\|^2$ and the original shift sum, directly yielding the claimed identity.

  1. The definitions of the energy $E$, the vectors $z$ and $z'$, and the flipped sign configuration $\varepsilon'$ are unfolded in the goal.

    The tactic `dsimp only` expands the local `let`-bindings, expressing the energy difference $E(\varepsilon, z) - E(\varepsilon', z')$ directly in terms of the vector norms and the linear coordinate sums.

    line 13

      dsimp only
  2. The linear shift sum under the flipped sign configuration $\varepsilon'$ satisfies $\sum_j \varepsilon'_j w_j s_j = \sum_j \varepsilon_j w_j s_j - 2 \varepsilon_i w_i s_i$.

    To compute the difference between the two sums, `Finset.sum_sub_distrib` distributes the subtraction into each term. For any index $b \neq i$, the updated sign $\varepsilon'_b$ matches $\varepsilon_b$ by `dif_neg`, so the difference vanishes by `sub_self`. By `Finset.sum_eq_single_of_mem` and `Finset.mem_univ`, the sum collapses to the single term at index $i$, where the difference evaluates to $(-\varepsilon_i - \varepsilon_i) w_i s_i = -2 \varepsilon_i w_i s_i$. Rearranging this difference establishes the claimed formula for the new sum.

    lines 14–26

    uses Finset.sum_sub_distrib, dif_neg, sub_self, Finset.sum_eq_single_of_mem, Finset.mem_univ

      have hsum : (∑ j, Function.update ε i (-ε i) j * w j * s j)
          = (∑ j, ε j * w j * s j) - 2 * (ε i * w i * s i) := by
        have key : (∑ j, Function.update ε i (-ε i) j * w j * s j) - (∑ j, ε j * w j * s j)
            = - 2 * (ε i * w i * s i) := by
          rw [← Finset.sum_sub_distrib]
          have hzero : ∀ b ∈ Finset.univ, b ≠ i →
              (Function.update ε i (-ε i) b * w b * s b - ε b * w b * s b) = 0 := by
            intro b hb hne
            simp only [Function.update, dif_neg hne, sub_self]
          rw [Finset.sum_eq_single_of_mem i (Finset.mem_univ i) hzero]
          simp [Function.update]
          ring
        linarith
  3. $\varepsilon_i^2 = 1$.

    Since $\varepsilon_i$ is either $1$ or $-1$ by hypothesis, squaring either value gives $1$.

    line 27

      have heps : ε i ^ 2 = 1 := by rcases hε with h | h <;> rw [h] <;> norm_num
  4. $(\varepsilon_i w_i)^2 = w_i^2$.

    Expanding $(\varepsilon_i w_i)^2 = \varepsilon_i^2 w_i^2$ and substituting $\varepsilon_i^2 = 1$ leaves $w_i^2$.

    lines 28–30

      have hsq : (ε i * w i) ^ 2 = w i ^ 2 := by
        have : (ε i * w i) ^ 2 = ε i ^ 2 * w i ^ 2 := by ring
        rw [this, heps]; ring
  5. The squared norm $\|z'\|^2$ and the flipped linear sum in $E(\varepsilon', z')$ are rewritten in terms of $z$, $\langle n_i, z \rangle$, and the original linear sum.

    Applying `plank_signed_sum_flip` expresses $z'$ as $z - (\varepsilon_i w_i) n_i$. Because $n_i$ is a unit vector, expanding the norm $\|z - (\varepsilon_i w_i) n_i\|^2$ using `norm_sub_smul_sq` yields $\|z\|^2 - 2 \varepsilon_i w_i \langle n_i, z \rangle + (\varepsilon_i w_i)^2$. Rewriting with `hsum` replaces the sum $\sum_j \varepsilon'_j w_j s_j$ with $\sum_j \varepsilon_j w_j s_j - 2 \varepsilon_i w_i s_i$.

    lines 31–33

    uses plank_signed_sum_flip, norm_sub_smul_sq

      rw [plank_signed_sum_flip n w ε i]
      rw [norm_sub_smul_sq _ (n i) (hn i) (ε i * w i)]
      rw [hsum]
  6. The energy difference $E(\varepsilon, z) - E(\varepsilon', z')$ equals $2 w_i (\varepsilon_i (\langle n_i, z \rangle - s_i) - w_i / 2)$.

    The squared norms $\|z\|^2$ and original linear sums $\sum_j \varepsilon_j w_j s_j$ cancel between $E(\varepsilon, z)$ and $E(\varepsilon', z')$. Substituting $(\varepsilon_i w_i)^2 = w_i^2$, both sides reduce to the same polynomial expression $2 \varepsilon_i w_i \langle n_i, z \rangle - 2 \varepsilon_i w_i s_i - w_i^2$, which is verified by non-linear arithmetic.

    line 34

      nlinarith [hsq]
Finset.sum_sub_distrib
The sum of a difference of two functions over a finite set equals the difference of their sums: $\sum_{x \in s} (f(x) - g(x)) = \sum_{x \in s} f(x) - \sum_{x \in s} g(x)$.
dif_neg
In a dependent if-then-else expression, when the condition is false, the expression evaluates to the false branch.
sub_self
Subtracting an element from itself gives zero: $a - a = 0$.
Finset.sum_eq_single_of_mem
If the terms of a sum over a finite set are all zero except possibly at a single element $a$, the sum equals the term at $a$.
Finset.mem_univ
Every element of a finite type belongs to the universal finset `Finset.univ`.
plank_signed_sum_flip
Flipping the sign $\varepsilon_i$ to $-\varepsilon_i$ in the scaled sum $\frac{1}{2} \sum_j (\varepsilon_j w_j) n_j$ subtracts $(\varepsilon_i w_i) n_i$ from the vector.
norm_sub_smul_sq
For any vector $z$ and unit vector $n$, the squared norm $\|z - a n\|^2$ expands as $\|z\|^2 - 2 a \langle n, z \rangle + a^2$.

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/-- Energy difference under flipping a single sign in a plank system. -/lemma plank_energy_diff_single_flip    {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V]    {ι : Type*} [Fintype ι] [DecidableEq ι]    (n : ι → V) (hn : ∀ i, ‖n i‖ = 1)    (w : ι → ℝ) (s : ι → ℝ) (ε : ι → ℝ) (i : ι) (hε : ε i = 1 ∨ ε i = -1) :    let ε' := Function.update ε i (-ε i)    let z := (1 / 2 : ℝ) • ∑ j, (ε j * w j) • n j    let z' := (1 / 2 : ℝ) • ∑ j, (ε' j * w j) • n j    let E := fun (sgn : ι → ℝ) (pt : V) => ‖pt‖ ^ 2 - ∑ j, sgn j * w j * s j    E ε z - E ε' z' = 2 * w i * (ε i * (⟪n i, z⟫ - s i) - w i / 2) := by  dsimp only  have hsum : (∑ j, Function.update ε i (-ε i) j * w j * s j)      = (∑ j, ε j * w j * s j) - 2 * (ε i * w i * s i) := by    have key : (∑ j, Function.update ε i (-ε i) j * w j * s j) - (∑ j, ε j * w j * s j)        = - 2 * (ε i * w i * s i) := by      rw [← Finset.sum_sub_distrib]      have hzero : ∀ b ∈ Finset.univ, b ≠ i →          (Function.update ε i (-ε i) b * w b * s b - ε b * w b * s b) = 0 := by        intro b hb hne        simp only [Function.update, dif_neg hne, sub_self]      rw [Finset.sum_eq_single_of_mem i (Finset.mem_univ i) hzero]      simp [Function.update]      ring    linarith  have heps : ε i ^ 2 = 1 := by rcases hε with h | h <;> rw [h] <;> norm_num  have hsq : (ε i * w i) ^ 2 = w i ^ 2 := by    have : (ε i * w i) ^ 2 = ε i ^ 2 * w i ^ 2 := by ring    rw [this, heps]; ring  rw [plank_signed_sum_flip n w ε i]  rw [norm_sub_smul_sq _ (n i) (hn i) (ε i * w i)]  rw [hsum]  nlinarith [hsq]
See it in the graphDiscuss this resultChallenge it