AFTD Auto-Formalizing Theoretical Domains

linear_flattening_testable_one_iff_zero

theoremverified

At dimension n = 1 the secant variety σ_r((ℂ¹)⊗³) is testable by the linear flattening method if and only if r = 0, i.e. the supremum of the potency over all linear flattenings on (ℂ¹)⊗³ is exactly 1.

Statement

theorem linear_flattening_testable_one_iff_zero (r : ℕ) :
    linear_flattening_testable 1 r ↔ r = 0
Source
step towards linear_flattening_testable_even_step
Verified
23 Oct 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 4
  • linear_flattening_testable_zero_orderFor every n ≥ 1, σ_0((ℂⁿ)⊗³) is testable by the linear flattening method (some linear flattening has potency greater than 0), so the…
  • linear_flattening_testableThe secant variety σ_r((ℂⁿ)⊗³) is testable by the linear flattening method if there exists a linear flattening L : (ℂⁿ)⊗³ → Mat_{u×v}(ℂ)…
  • linear_flattening_potency_gtA linear flattening L : ℂⁿ ⊗ ℂⁿ ⊗ ℂⁿ → Mat_{u×v}(ℂ) has potency greater than r: some tensor T has rank L(T) > r · ρ(L), where ρ(L) =…
  • tensor3_pureThe pure tensor x ⊗ y ⊗ z in ℂⁿ ⊗ ℂⁿ ⊗ ℂⁿ, with entries x_i y_j z_k.

The proof, explained

At dimension n = 1, the tensor space (ℂ¹)⊗³ is one-dimensional and spanned by the pure tensor 1 ⊗ 1 ⊗ 1, so every tensor T is a scalar multiple of 1 ⊗ 1 ⊗ 1. Consequently, any linear flattening L maps T to a scalar multiple of L(1 ⊗ 1 ⊗ 1), which cannot have strictly larger rank than L(1 ⊗ 1 ⊗ 1). Assuming r ≥ 1 would require r · rank(L(1 ⊗ 1 ⊗ 1)) < rank(L(T)), a contradiction; hence only r = 0 can be testable, and order 0 is testable by a general base lemma.

  1. To prove the forward direction by contradiction, assume r ≥ 1 and specialize the testability condition to the pure tensor 1 ⊗ 1 ⊗ 1.

    The `constructor` separates the equivalence into forward and reverse implications. For the forward direction, testability of order r provides dimensions u, v, a linear flattening L, and a tensor T satisfying r * rank(L(x ⊗ y ⊗ z)) < rank(L(T)) for all x, y, z. Supposing r ≠ 0, we have 1 ≤ r via `Nat.one_le_iff_ne_zero.mpr`. Evaluating the testability condition at x = y = z = 1 gives r * rank(L(1 ⊗ 1 ⊗ 1)) < rank(L(T)).

    lines 4–9

    uses Nat.one_le_iff_ne_zero.mpr

      constructor
      · intro h
        rcases h with ⟨u, v, L, T, hT⟩
        by_contra hr
        have hr1 : 1 ≤ r := Nat.one_le_iff_ne_zero.mpr hr
        have hT_spec : r * (L (tensor3_pure 1 1 1)).rank < (L T).rank := hT 1 1 1
  2. The flattened tensor L(T) is a scalar multiple of L(1 ⊗ 1 ⊗ 1), specifically L(T) = T(0,0,0) • L(1 ⊗ 1 ⊗ 1).

    Since the index set Fin 1 is a subsingleton, any indices i, j, k equal 0 by `Subsingleton.elim`. Extensionality and simplification show that T = T(0,0,0) • (1 ⊗ 1 ⊗ 1). Applying the linear map L and using linearity (`map_smul`) yields L(T) = T(0,0,0) • L(1 ⊗ 1 ⊗ 1).

    lines 10–15

    uses Subsingleton.elim, map_smul

        have hLT : L T = (T 0 0 0) • L (tensor3_pure 1 1 1) := by
          have hT_eq : T = (T 0 0 0) • tensor3_pure 1 1 1 := by
            ext i j k
            simp [tensor3_pure, Subsingleton.elim i 0, Subsingleton.elim j 0, Subsingleton.elim k 0]
          conv_lhs => rw [hT_eq]
          rw [map_smul]
  3. The rank of L(T) is at most the rank of L(1 ⊗ 1 ⊗ 1).

    Using the identity L(T) = T(0,0,0) • L(1 ⊗ 1 ⊗ 1), we case-split on whether T(0,0,0) = 0: if it is zero, L(T) is the zero matrix (`zero_smul`), which has rank 0 (`Matrix.rank_zero`) and is thus bounded by `Nat.zero_le`; if it is non-zero, it is a non-zero divisor (`mem_nonZeroDivisors_of_ne_zero`), so scaling preserves the matrix rank (`Matrix.rank_smul_of_mem_nonZeroDivisors`), giving rank(L(T)) = rank(L(1 ⊗ 1 ⊗ 1)). In either case, rank(L(T)) ≤ rank(L(1 ⊗ 1 ⊗ 1)).

    lines 16–22

    uses zero_smul, Matrix.rank_zero, Nat.zero_le, mem_nonZeroDivisors_of_ne_zero, Matrix.rank_smul_of_mem_nonZeroDivisors

        have h_rank_le : (L T).rank ≤ (L (tensor3_pure 1 1 1)).rank := by
          rw [hLT]
          by_cases hc : T 0 0 0 = 0
          · rw [hc, zero_smul, Matrix.rank_zero]
            exact Nat.zero_le _
          · have h_mem : T 0 0 0 ∈ nonZeroDivisors ℂ := mem_nonZeroDivisors_of_ne_zero hc
            rw [Matrix.rank_smul_of_mem_nonZeroDivisors (L (tensor3_pure 1 1 1)) h_mem]
  4. The chain of inequalities produces a contradiction, showing that r must be 0.

    Since 1 ≤ r, we have rank(L(1 ⊗ 1 ⊗ 1)) = 1 * rank(L(1 ⊗ 1 ⊗ 1)) ≤ r * rank(L(1 ⊗ 1 ⊗ 1)) by `one_mul` and `Nat.mul_le_mul_right`. Combining this with rank(L(T)) ≤ rank(L(1 ⊗ 1 ⊗ 1)) and r * rank(L(1 ⊗ 1 ⊗ 1)) < rank(L(T)) yields a strict cycle of inequalities, which `omega` uses to derive a contradiction (False).

    lines 23–26

    uses one_mul, Nat.mul_le_mul_right

        have h_le : (L (tensor3_pure 1 1 1)).rank ≤ r * (L (tensor3_pure 1 1 1)).rank := by
          calc (L (tensor3_pure 1 1 1)).rank = 1 * (L (tensor3_pure 1 1 1)).rank := by rw [one_mul]
          _ ≤ r * (L (tensor3_pure 1 1 1)).rank := Nat.mul_le_mul_right (L (tensor3_pure 1 1 1)).rank hr1
        omega
  5. Order r = 0 is testable.

    For the reverse direction, we assume r = 0 and invoke the existing theorem `linear_flattening_testable_zero_order` at dimension 1, using reflexivity `le_rfl` to supply the hypothesis 1 ≤ 1.

    lines 27–28

    uses linear_flattening_testable_zero_order, le_rfl

      · rintro rfl
        exact linear_flattening_testable_zero_order 1 le_rfl
Nat.one_le_iff_ne_zero.mpr
a natural number is at least 1 if it is not 0
Subsingleton.elim
in a subsingleton type, any two elements are equal
map_smul
a linear map preserves scalar multiplication: L(c • x) = c • L(x)
zero_smul
scaling any element by 0 gives 0
Matrix.rank_zero
the rank of the zero matrix is 0
Nat.zero_le
0 is less than or equal to any natural number
mem_nonZeroDivisors_of_ne_zero
in a domain with no zero divisors, every non-zero element is a non-zero divisor
Matrix.rank_smul_of_mem_nonZeroDivisors
scaling a matrix by a non-zero divisor preserves its rank
one_mul
multiplying by 1 on the left is the identity: 1 * a = a
Nat.mul_le_mul_right
multiplication of natural numbers preserves inequality: if n ≤ m, then n * k ≤ m * k
linear_flattening_testable_zero_order
for any dimension n ≥ 1, order 0 is testable by the linear flattening method
le_rfl
the order relation is reflexive: 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

/-- Exact small instance: at dimension 1 every linear flattening has potency at most 1, so the only testable order is 0. -/theorem linear_flattening_testable_one_iff_zero (r : ℕ) :    linear_flattening_testable 1 r ↔ r = 0 := by  constructor  · intro h    rcases h with ⟨u, v, L, T, hT⟩    by_contra hr    have hr1 : 1 ≤ r := Nat.one_le_iff_ne_zero.mpr hr    have hT_spec : r * (L (tensor3_pure 1 1 1)).rank < (L T).rank := hT 1 1 1    have hLT : L T = (T 0 0 0) • L (tensor3_pure 1 1 1) := by      have hT_eq : T = (T 0 0 0) • tensor3_pure 1 1 1 := by        ext i j k        simp [tensor3_pure, Subsingleton.elim i 0, Subsingleton.elim j 0, Subsingleton.elim k 0]      conv_lhs => rw [hT_eq]      rw [map_smul]    have h_rank_le : (L T).rank ≤ (L (tensor3_pure 1 1 1)).rank := by      rw [hLT]      by_cases hc : T 0 0 0 = 0      · rw [hc, zero_smul, Matrix.rank_zero]        exact Nat.zero_le _      · have h_mem : T 0 0 0 ∈ nonZeroDivisors ℂ := mem_nonZeroDivisors_of_ne_zero hc        rw [Matrix.rank_smul_of_mem_nonZeroDivisors (L (tensor3_pure 1 1 1)) h_mem]    have h_le : (L (tensor3_pure 1 1 1)).rank ≤ r * (L (tensor3_pure 1 1 1)).rank := by      calc (L (tensor3_pure 1 1 1)).rank = 1 * (L (tensor3_pure 1 1 1)).rank := by rw [one_mul]      _ ≤ r * (L (tensor3_pure 1 1 1)).rank := Nat.mul_le_mul_right (L (tensor3_pure 1 1 1)).rank hr1    omega  · rintro rfl    exact linear_flattening_testable_zero_order 1 le_rfl
See it in the graphDiscuss this resultChallenge it