real_multiplicative_defect_deletion_contraction_le
Let a_D, a_star_D, t_D, a_Q, a_star_Q, t_Q be non-negative real numbers. If t_D ^ 2 ≤ a_D * a_star_D and t_Q ^ 2 ≤ a_Q * a_star_Q, then (t_D + t_Q) ^ 2 ≤ (a_D + a_Q) * (a_star_D + a_star_Q).
Statement
theorem real_multiplicative_defect_deletion_contraction_le
{a_D a_star_D t_D a_Q a_star_Q t_Q : ℝ}
(ha_D : 0 ≤ a_D) (ha_star_D : 0 ≤ a_star_D) (ht_D : 0 ≤ t_D)
(ha_Q : 0 ≤ a_Q) (ha_star_Q : 0 ≤ a_star_Q) (ht_Q : 0 ≤ t_Q)
(hD : t_D ^ 2 ≤ a_D * a_star_D) (hQ : t_Q ^ 2 ≤ a_Q * a_star_Q) :
(t_D + t_Q) ^ 2 ≤ (a_D + a_Q) * (a_star_D + a_star_Q)- Source
- arXiv:2610.27706 Lemma 5.2
- Verified
- 22 Oct 2026
- Axioms
Classical.choiceQuot.soundpropext
Read back from the Lean
For all real numbers $a_D, a^*_D, t_D, a_Q, a^*_Q, t_Q \in \mathbb{R}$, if $a_D \ge 0$, $a^*_D \ge 0$, $t_D \ge 0$, $a_Q \ge 0$, $a^*_Q \ge 0$, and $t_Q \ge 0$, and if $t_D^2 \le a_D \cdot a^*_D$ and $t_Q^2 \le a_Q \cdot a^*_Q$, then $(t_D + t_Q)^2 \le (a_D + a_Q)(a^*_D + a^*_Q)$.
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
To prove $(t_D + t_Q)^2 \le (a_D + a_Q)(a^*_D + a^*_Q)$, the argument expands both sides into square terms and a cross term. Since the squares are bounded by assumption ($t_D^2 \le a_D a^*_D$ and $t_Q^2 \le a_Q a^*_Q$), it suffices to bound the cross term $2 t_D t_Q \le a_D a^*_Q + a_Q a^*_D$. This is done by bounding its square via the AM-GM identity $4uv \le (u + v)^2$ and taking square roots using the non-negativity of the variables.
$(2 t_D t_Q)^2 \le 4 (a_D a^*_D) (a_Q a^*_Q)$
Multiplying the two given hypotheses $t_D^2 \le a_D a^*_D$ and $t_Q^2 \le a_Q a^*_Q$ together and scaling by 4 gives $4 t_D^2 t_Q^2 \le 4 (a_D a^*_D)(a_Q a^*_Q)$, which is $(2 t_D t_Q)^2 \le 4 (a_D a^*_D)(a_Q a^*_Q)$. The `nlinarith` tactic closes this non-linear arithmetic step automatically from the non-negativity of the variables.
have h1 : (2 * t_D * t_Q) ^ 2 ≤ 4 * (a_D * a_star_D) * (a_Q * a_star_Q) := by nlinarith
$4 (a_D a^*_D) (a_Q a^*_Q) \le (a_D a^*_Q + a_Q a^*_D)^2$
By `sq_nonneg`, the square $(a_D a^*_Q - a_Q a^*_D)^2$ is non-negative. Since $(u + v)^2 - 4uv = (u - v)^2 \ge 0$, setting $u = a_D a^*_Q$ and $v = a_Q a^*_D$ implies $4 (a_D a^*_D)(a_Q a^*_Q) \le (a_D a^*_Q + a_Q a^*_D)^2$, which `linarith` verifies.
uses
sq_nonneghave h2 : 4 * (a_D * a_star_D) * (a_Q * a_star_Q) ≤ (a_D * a_star_Q + a_Q * a_star_D) ^ 2 := by have : 0 ≤ (a_D * a_star_Q - a_Q * a_star_D) ^ 2 := sq_nonneg _ linarith
$(2 t_D t_Q)^2 \le (a_D a^*_Q + a_Q a^*_D)^2$
Combining the previous two bounds ($h_1$ and $h_2$) by transitivity via `linarith` gives the bound on the squared cross term.
have h_sq : (2 * t_D * t_Q) ^ 2 ≤ (a_D * a_star_Q + a_Q * a_star_D) ^ 2 := by linarith
Both $2 t_D t_Q$ and $a_D a^*_Q + a_Q a^*_D$ are non-negative.
Because all original variables are non-negative, their products and sums are non-negative as well, as established by the `positivity` tactic.
have h_cross_nonneg : 0 ≤ 2 * t_D * t_Q := by positivity have h_sum_nonneg : 0 ≤ a_D * a_star_Q + a_Q * a_star_D := by positivity
$2 t_D t_Q \le a_D a^*_Q + a_Q a^*_D$
By `sq_le_sq`, $u^2 \le v^2$ is equivalent to $|u| \le |v|$. Since both terms are non-negative, `abs_of_nonneg` simplifies the absolute values to the terms themselves, reducing $(2 t_D t_Q)^2 \le (a_D a^*_Q + a_Q a^*_D)^2$ to $2 t_D t_Q \le a_D a^*_Q + a_Q a^*_D$.
uses
sq_le_sq,abs_of_nonneghave h_cross : 2 * t_D * t_Q ≤ a_D * a_star_Q + a_Q * a_star_D := by rw [sq_le_sq, abs_of_nonneg h_cross_nonneg, abs_of_nonneg h_sum_nonneg] at h_sq exact h_sq
$(t_D + t_Q)^2 \le (a_D + a_Q)(a^*_D + a^*_Q)$
Expanding both sides gives $(t_D + t_Q)^2 = t_D^2 + 2 t_D t_Q + t_Q^2$ and $(a_D + a_Q)(a^*_D + a^*_Q) = a_D a^*_D + (a_D a^*_Q + a_Q a^*_D) + a_Q a^*_Q$. Summing the given inequalities $t_D^2 \le a_D a^*_D$ ($h_D$) and $t_Q^2 \le a_Q a^*_Q$ ($h_Q$) with the cross-term bound $2 t_D t_Q \le a_D a^*_Q + a_Q a^*_D$ ($h_{cross}$), `linarith` concludes the proof.
linarith
sq_nonneg- In an ordered semiring, the square of any element is non-negative ($0 \le a^2$).
sq_le_sq- In a strictly ordered ring, $a^2 \le b^2$ holds if and only if $|a| \le |b|$.
abs_of_nonneg- The absolute value of a non-negative element is the element itself (if $0 \le a$, then $|a| = a$).
- The `linarith` tactic (lines 12 and 20) automatically handles the algebraic expansion of binomial squares and verifies the linear combinations of the inequalities.
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
Choose a tactic — a marked line — to see the goals Lean had before and after it, as in Lean’s infoview. Step with ← → or j k.
/-- AM-GM inequality step: non-negative real numbers satisfying multiplicative defect bounds preserve the bound under addition. -/theorem real_multiplicative_defect_deletion_contraction_le {a_D a_star_D t_D a_Q a_star_Q t_Q : ℝ} (ha_D : 0 ≤ a_D) (ha_star_D : 0 ≤ a_star_D) (ht_D : 0 ≤ t_D) (ha_Q : 0 ≤ a_Q) (ha_star_Q : 0 ≤ a_star_Q) (ht_Q : 0 ≤ t_Q) (hD : t_D ^ 2 ≤ a_D * a_star_D) (hQ : t_Q ^ 2 ≤ a_Q * a_star_Q) : (t_D + t_Q) ^ 2 ≤ (a_D + a_Q) * (a_star_D + a_star_Q) := by have h1 : (2 * t_D * t_Q) ^ 2 ≤ 4 * (a_D * a_star_D) * (a_Q * a_star_Q) := by nlinarith have h2 : 4 * (a_D * a_star_D) * (a_Q * a_star_Q) ≤ (a_D * a_star_Q + a_Q * a_star_D) ^ 2 := by have : 0 ≤ (a_D * a_star_Q - a_Q * a_star_D) ^ 2 := sq_nonneg _ linarith have h_sq : (2 * t_D * t_Q) ^ 2 ≤ (a_D * a_star_Q + a_Q * a_star_D) ^ 2 := by linarith have h_cross_nonneg : 0 ≤ 2 * t_D * t_Q := by positivity have h_sum_nonneg : 0 ≤ a_D * a_star_Q + a_Q * a_star_D := by positivity have h_cross : 2 * t_D * t_Q ≤ a_D * a_star_Q + a_Q * a_star_D := by rw [sq_le_sq, abs_of_nonneg h_cross_nonneg, abs_of_nonneg h_sum_nonneg] at h_sq exact h_sq linarith