AFTD Auto-Formalizing Theoretical Domains

matrix_row_sum_zero_shifted_eigenvector

theoremverified

For real order-four matrices with zero row sums, an eigenvector of one with nonzero eigenvalue can be adjusted by a constant to give an eigenvector of the other when its image differs from a nonzero scalar multiple by a constant.

Statement

theorem matrix_row_sum_zero_shifted_eigenvector
    (L K : Matrix (Fin 4) (Fin 4) ℝ)
    (hrL : ∀ i, ∑ j, L i j = 0) (hrK : ∀ i, ∑ j, K i j = 0)
    (v : Fin 4 → ℂ) (hv0 : v ≠ 0) (z : ℂ) (hz : z ≠ 0)
    (hv : (L.map (algebraMap ℝ ℂ)).mulVec v = z • v)
    (lam c : ℂ) (hlam : lam ≠ 0)
    (hKv : (K.map (algebraMap ℝ ℂ)).mulVec v = fun i => lam * v i + c) :
    ∃ w : Fin 4 → ℂ, w ≠ 0 ∧
      (K.map (algebraMap ℝ ℂ)).mulVec w = lam • w
Source
Sector-bound formalization for IN-166; arXiv:math/0508176, Theorem 5.
Verified
27 Oct 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 1
Used by 2

The proof, explained

The candidate eigenvector is constructed by shifting $v$ by a constant vector: $w = v + \frac{c}{\lambda}\mathbf{1}$. Because $K$ has zero row sums, it annihilates constant vectors, so linearity yields $K w = K v = \lambda v + c\mathbf{1} = \lambda w$. To show that $w \neq 0$, suppose for contradiction that $w = 0$; then $v = -\frac{c}{\lambda}\mathbf{1}$ is constant, which forces $L v = 0$ because $L$ also has zero row sums. This implies $z v = 0$, forcing $v = 0$ since $z \neq 0$, which contradicts the assumption that $v \neq 0$.

  1. Define the candidate eigenvector $w = v + \frac{c}{\lambda}\mathbf{1}$.

    We shift the given eigenvector $v$ by the constant vector whose entries all equal $c / \lambda$.

    line 10

      let w : Fin 4 → ℂ := v + fun _ => c / lam
  2. $w \neq 0$.

    Suppose, for contradiction, that $w = 0$. Evaluating at each coordinate $i$ via `congrFun` gives $v_i + c / \lambda = 0$, so $v$ is the constant vector $-c / \lambda \mathbf{1}$. Because $L$ has zero row sums, multiplying it by any constant vector yields zero (`matrix_order_four_row_sum_zero_mulVec_const`), which implies $L v = 0$. Since $v$ is an eigenvector of $L$ with eigenvalue $z$, we have $z \cdot v = L v = 0$. Since $z \neq 0$, the zero-product property for scalar multiplication (`smul_eq_zero.mp`, `resolve_left`) implies $v = 0$, contradicting the hypothesis $v \neq 0$.

    lines 11–22

    uses congrFun, matrix_order_four_row_sum_zero_mulVec_const, smul_eq_zero.mp, resolve_left

      have hw0 : w ≠ 0 := by
        intro hh
        have hvconst : v = fun _ => -(c / lam) := by
          ext i
          have hi := congrFun hh i
          change v i + c / lam = 0 at hi
          linear_combination hi
        have hzero : (L.map (algebraMap ℝ ℂ)).mulVec v = 0 := by
          rw [hvconst]
          exact matrix_order_four_row_sum_zero_mulVec_const L hrL _
        have hsm : z • v = 0 := hv.symm.trans hzero
        exact hv0 ((smul_eq_zero.mp hsm).resolve_left hz)
  3. It suffices to show $K w = \lambda w$, and $K w$ simplifies to $\lambda v + c \mathbf{1}$.

    We provide $w$ and the established fact $w \neq 0$ as the witness. By linearity of matrix-vector multiplication (`Matrix.mulVec_add`), $K w = K(v + \frac{c}{\lambda}\mathbf{1}) = K v + K(\frac{c}{\lambda}\mathbf{1})$. Because $K$ has zero row sums, $K(\frac{c}{\lambda}\mathbf{1}) = 0$ (`matrix_order_four_row_sum_zero_mulVec_const`). Using $a + 0 = a$ (`add_zero`) and the given image $K v = \lambda v + c \mathbf{1}$, the left-hand side reduces to $\lambda v + c \mathbf{1}$.

    lines 23–25

    uses Matrix.mulVec_add, matrix_order_four_row_sum_zero_mulVec_const, add_zero

      refine ⟨w, hw0, ?_⟩
      change (K.map (algebraMap ℝ ℂ)).mulVec (v + fun _ => c / lam) = _
      rw [Matrix.mulVec_add, matrix_order_four_row_sum_zero_mulVec_const K hrK, add_zero, hKv]
  4. $K w = \lambda w$, completing the proof.

    It remains to verify the vector equality componentwise: for each coordinate $i$, we must show $\lambda v_i + c = \lambda(v_i + c / \lambda)$. Clearing the denominator using $\lambda \neq 0$ and distributing $\lambda$ across the sum confirms that both sides are identical.

    lines 26–29

      ext i
      change lam * v i + c = lam * (v i + c / lam)
      field_simp
      <;> ring
congrFun
if two functions are equal, their values at any given input are equal
matrix_order_four_row_sum_zero_mulVec_const
a real 4 × 4 matrix whose row sums are all zero annihilates any constant complex vector
smul_eq_zero.mp
if a scalar multiple $c \cdot x = 0$, then $c = 0$ or $x = 0$
resolve_left
if a disjunction $a \lor b$ holds and $a$ is false, then $b$ holds
Matrix.mulVec_add
matrix-vector multiplication distributes over vector addition: $A(x + y) = Ax + Ay$
add_zero
adding zero to an element leaves it unchanged: $a + 0 = a$

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

theorem matrix_row_sum_zero_shifted_eigenvector    (L K : Matrix (Fin 4) (Fin 4) ℝ)    (hrL : ∀ i, ∑ j, L i j = 0) (hrK : ∀ i, ∑ j, K i j = 0)    (v : Fin 4 → ℂ) (hv0 : v ≠ 0) (z : ℂ) (hz : z ≠ 0)    (hv : (L.map (algebraMap ℝ ℂ)).mulVec v = z • v)    (lam c : ℂ) (hlam : lam ≠ 0)    (hKv : (K.map (algebraMap ℝ ℂ)).mulVec v = fun i => lam * v i + c) :    ∃ w : Fin 4 → ℂ, w ≠ 0 ∧      (K.map (algebraMap ℝ ℂ)).mulVec w = lam • w := by  let w : Fin 4 → ℂ := v + fun _ => c / lam  have hw0 : w ≠ 0 := by    intro hh    have hvconst : v = fun _ => -(c / lam) := by      ext i      have hi := congrFun hh i      change v i + c / lam = 0 at hi      linear_combination hi    have hzero : (L.map (algebraMap ℝ ℂ)).mulVec v = 0 := by      rw [hvconst]      exact matrix_order_four_row_sum_zero_mulVec_const L hrL _    have hsm : z • v = 0 := hv.symm.trans hzero    exact hv0 ((smul_eq_zero.mp hsm).resolve_left hz)  refine ⟨w, hw0, ?_⟩  change (K.map (algebraMap ℝ ℂ)).mulVec (v + fun _ => c / lam) = _  rw [Matrix.mulVec_add, matrix_order_four_row_sum_zero_mulVec_const K hrK, add_zero, hKv]  ext i  change lam * v i + c = lam * (v i + c / lam)  field_simp  <;> ring
See it in the graphDiscuss this resultChallenge it