tournament_degree_prod_le_regular
For n ≥ 2 and any real numbers x_1, ..., x_n ≥ 1 summing to n^2, the product ∏_{i=1}^n (x_i^2 + n) is at most (n(n+1))^n.
Statement
theorem tournament_degree_prod_le_regular (n : ℕ) (hn : 2 ≤ n) (x : Fin n → ℝ) (hx_pos : ∀ i, 1 ≤ x i) (hx_sum : ∑ i : Fin n, x i = (n : ℝ) ^ 2) : ∏ i : Fin n, (x i ^ 2 + (n : ℝ)) ≤ ((n : ℝ) * ((n : ℝ) + 1)) ^ n- Source
- arXiv:2610.25886 Proposition 4.1 proof
- Verified
- 22 Oct 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 4
- ProbabilisticTheory.OrderedVectorSpaceAn ordered real vector space.
- ProbabilisticTheory.instArchimedeanOrderUnitSpaceRealThe real numbers form an Archimedean order-unit space.
- ProbabilisticTheory.OrderUnitSpaceAn ordered real vector space whose distinguished element
1is an order unit: it is nonnegative, and every element is bounded above by a… - ProbabilisticTheory.ArchimedeanOrderUnitSpaceAn order-unit space whose distinguished order unit is Archimedean.
Read back from the Lean
For every natural number $n$ such that $2 \le n$, and for every function $x : \text{Fin } n \to \mathbb{R}$, if $1 \le x(i)$ for all $i \in \text{Fin } n$ and $\sum_{i \in \text{Fin } n} x(i) = n^2$, then $\prod_{i \in \text{Fin } n} (x(i)^2 + n) \le (n(n + 1))^n$.
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 theorem is proved by taking logarithms and applying a tangent-line upper bound. An auxiliary lemma shows that for any m ≥ 2 and x ≥ 1, ln(x² + m) ≤ ln(m(m + 1)) + 2(x - m)/(m + 1), by demonstrating that the function φ(y) = ln(y² + m) - 2y/(m + 1) increases on [1, m] and decreases on [m, ∞), reaching its maximum at y = m. Setting m = n and summing this inequality over all x_i, the linear perturbation terms cancel because ∑ x_i = n². Exponentiating the resulting upper bound on the sum of logarithms yields the desired product bound ∏ (x_i² + n) ≤ (n(n + 1))^n.
Define the auxiliary function φ(y) = ln(y² + m) - 2y/(m + 1) and establish that it is continuous on ℝ.
Since m ≥ 2, y² + m ≥ m > 0 for all y, so y² + m is never zero. The composition y ↦ ln(y² + m) is therefore continuous on ℝ, and subtracting the linear term 2y/(m + 1) preserves continuity.
uses
Continuous.log,Continuous.sub,sq_nonneglemma tournament_degree_prod_le_regular_aux (m : ℝ) (hm : 2 ≤ m) (x : ℝ) (hx : 1 ≤ x) : Real.log (x ^ 2 + m) ≤ Real.log (m * (m + 1)) + 2 * (x - m) / (m + 1) := by have hm1 : 0 < m + 1 := by linarith let φ : ℝ → ℝ := fun y => Real.log (y ^ 2 + m) - 2 * y / (m + 1) have hcont_log : Continuous (fun y : ℝ => Real.log (y ^ 2 + m)) := by apply Continuous.log · fun_prop · intro y have : 0 < y ^ 2 + m := by nlinarith [sq_nonneg y] exact this.ne' have hcont : Continuous φ := by apply Continuous.sub hcont_log fun_prop
For every y ∈ ℝ, the derivative of φ at y is φ'(y) = 2y/(y² + m) - 2/(m + 1).
By the power rule and linearity, the derivative of z ↦ z² + m is 2y. Applying the chain rule to the logarithm yields 2y/(y² + m). Differentiating the linear term 2z/(m + 1) gives 2/(m + 1). Subtracting the two gives the derivative of φ.
uses
sq_nonneg,hasDerivAt_pow,hasDerivAt_add_const_iff,hasDerivAt_const_mul,div_eq_mul_inv,mul_comm,mul_left_comm,mul_assochave hderiv : ∀ y : ℝ, HasDerivAt φ (2 * y / (y ^ 2 + m) - 2 / (m + 1)) y := by intro y have hy : y ^ 2 + m ≠ 0 := by have : 0 < y ^ 2 + m := by nlinarith [sq_nonneg y] exact this.ne' have h1 : HasDerivAt (fun z : ℝ => z ^ 2 + m) (2 * y) y := by have hbase : HasDerivAt (fun z : ℝ => z ^ 2) (2 * y) y := by simpa using hasDerivAt_pow 2 y exact (hasDerivAt_add_const_iff m).mpr hbase have h2 : HasDerivAt (fun z : ℝ => Real.log (z ^ 2 + m)) (2 * y / (y ^ 2 + m)) y := by simpa using h1.log hy have h3 : HasDerivAt (fun z : ℝ => 2 * z / (m + 1)) (2 / (m + 1)) y := by have h := hasDerivAt_const_mul (x := y) (c := 2 / (m + 1)) simpa [div_eq_mul_inv, mul_comm, mul_left_comm, mul_assoc] using h have hsub := h2.sub h3 change HasDerivAt (fun z : ℝ => Real.log (z ^ 2 + m) - 2 * z / (m + 1)) (2 * y / (y ^ 2 + m) - 2 / (m + 1)) y at hsub simpa [φ] using hsub
The function φ is monotone increasing on the closed interval [1, m].
By the mean value theorem / derivative test on the convex interval [1, m], it suffices to show φ'(y) ≥ 0 on the interior (1, m). On this interval, 1 < y < m, so 2y(m + 1) - 2(y² + m) = 2(y - 1)(m - y) ≥ 0, which implies 2y/(y² + m) ≥ 2/(m + 1).
uses
monotoneOn_of_hasDerivWithinAt_nonneg,convex_Icc,interior_Icc,sq_nonneg,sub_nonneg,div_le_div_iff₀,mul_nonneghave hmono : MonotoneOn φ (Set.Icc 1 m) := by apply monotoneOn_of_hasDerivWithinAt_nonneg (convex_Icc (1:ℝ) m) hcont.continuousOn · intro y hy exact (hderiv y).hasDerivWithinAt · intro y hy have hy' : y ∈ Set.Ioo (1:ℝ) m := by simpa [interior_Icc] using hy have hy1 : 1 < y := hy'.1 have hym : y < m := hy'.2 have hden : 0 < y ^ 2 + m := by nlinarith [sq_nonneg y] have hden2 : 0 < m + 1 := hm1 rw [sub_nonneg] exact (div_le_div_iff₀ hden2 hden).mpr (by nlinarith [mul_nonneg (sub_nonneg.mpr hy1.le) (sub_nonneg.mpr hym.le)])
The function φ is antitone (decreasing) on the interval [m, ∞).
On the interior (m, ∞), we have y > m ≥ 2 > 1, which implies 2(y - 1)(m - y) ≤ 0, so φ'(y) ≤ 0. Since [m, ∞) is convex and φ is continuous, φ is antitone on [m, ∞).
uses
antitoneOn_of_hasDerivWithinAt_nonpos,convex_Ici,interior_Ici,sq_nonneg,sub_nonpos,div_le_div_iff₀,mul_nonneghave hanti : AntitoneOn φ (Set.Ici m) := by apply antitoneOn_of_hasDerivWithinAt_nonpos (convex_Ici m) hcont.continuousOn · intro y hy exact (hderiv y).hasDerivWithinAt · intro y hy have hy' : y ∈ Set.Ioi m := by simpa [interior_Ici] using hy have hym : m < y := hy' have hden : 0 < y ^ 2 + m := by nlinarith [sq_nonneg y] have hden2 : 0 < m + 1 := hm1 rw [sub_nonpos] exact (div_le_div_iff₀ hden hden2).mpr (by nlinarith [mul_nonneg (sub_nonneg.mpr (by linarith : 1 ≤ y)) (sub_nonneg.mpr hym.le)])
For any x ≥ 1, ln(x² + m) ≤ ln(m(m + 1)) + 2(x - m)/(m + 1).
We split into two cases: if x ≤ m, then x ∈ [1, m], and monotonicity of φ gives φ(x) ≤ φ(m). If x > m, then x ∈ [m, ∞), and antitonicity gives φ(x) ≤ φ(m). In either case, φ(x) ≤ φ(m), which expands to ln(x² + m) - 2x/(m + 1) ≤ ln(m² + m) - 2m/(m + 1). Factoring m² + m = m(m + 1) and rearranging yields the desired inequality.
uses
le_rflby_cases hxle : x ≤ m · have hxI : x ∈ Set.Icc (1:ℝ) m := Set.mem_Icc.mpr ⟨hx, hxle⟩ have hmI : m ∈ Set.Icc (1:ℝ) m := Set.mem_Icc.mpr ⟨by linarith, le_rfl⟩ have h := hmono hxI hmI hxle simp only [φ] at h have hφm : m ^ 2 + m = m * (m + 1) := by ring rw [hφm] at h have hsub : 2 * (x - m) / (m + 1) = 2 * x / (m + 1) - 2 * m / (m + 1) := by ring rw [hsub] linarith · have hxge : m ≤ x := by linarith have hxI : x ∈ Set.Ici m := Set.mem_Ici.mpr hxge have hmI : m ∈ Set.Ici m := Set.mem_Ici.mpr le_rfl have h := hanti hmI hxI hxge simp only [φ] at h have hφm : m ^ 2 + m = m * (m + 1) := by ring rw [hφm] at h have hsub : 2 * (x - m) / (m + 1) = 2 * x / (m + 1) - 2 * m / (m + 1) := by ring rw [hsub] linarith
Let N = n. Then N ≥ 2, N > 0, N + 1 > 0, and x_i² + N > 0 for each i.
We cast the integer n to a real number N. Since n ≥ 2, N ≥ 2 > 0. For each index i, x_i ≥ 1, so x_i² + N ≥ 1 + 2 > 0.
uses
sq_nonneg/-- Degree product bound for tournament arborescences from concavity. -/ theorem tournament_degree_prod_le_regular (n : ℕ) (hn : 2 ≤ n) (x : Fin n → ℝ) (hx_pos : ∀ i, 1 ≤ x i) (hx_sum : ∑ i : Fin n, x i = (n : ℝ) ^ 2) : ∏ i : Fin n, (x i ^ 2 + (n : ℝ)) ≤ ((n : ℝ) * ((n : ℝ) + 1)) ^ n := by let N : ℝ := (n : ℝ) have hN2 : 2 ≤ N := by dsimp [N] exact_mod_cast hn have hN0 : 0 < N := by linarith have hN1 : 0 < N + 1 := by linarith have hpos : ∀ i : Fin n, 0 < x i ^ 2 + N := by intro i have hxi : 1 ≤ x i := hx_pos i nlinarith [sq_nonneg (x i)]
For each i, ln(x_i² + N) ≤ ln(N(N + 1)) + 2(x_i - N)/(N + 1).
This is an immediate application of the auxiliary lemma with m = N and x = x_i, since N ≥ 2 and x_i ≥ 1.
uses
tournament_degree_prod_le_regular_auxhave hpoint : ∀ i : Fin n, Real.log (x i ^ 2 + N) ≤ Real.log (N * (N + 1)) + 2 * (x i - N) / (N + 1) := by intro i exact tournament_degree_prod_le_regular_aux N hN2 (x i) (hx_pos i)
The sum of logarithms ∑_{i} ln(x_i² + N) is at most ∑_{i} (ln(N(N + 1)) + 2(x_i - N)/(N + 1)).
Summing the pointwise upper bounds over all i ∈ Fin n preserves the inequality.
uses
Finset.sum_le_sumhave hsum_le : ∑ i : Fin n, Real.log (x i ^ 2 + N) ≤ ∑ i : Fin n, (Real.log (N * (N + 1)) + 2 * (x i - N) / (N + 1)) := by exact Finset.sum_le_sum (fun i _ => hpoint i)
The right-hand sum simplifies to n * ln(N(N + 1)).
The sum splits into a constant sum and a linear deviation sum. The constant term sums to n * ln(N(N + 1)). In the deviation term, the factor 2/(N + 1) pulls out, leaving ∑ (x_i - N) = ∑ x_i - n * N = N² - n * N = 0, because ∑ x_i = n² and N = n. Thus the linear term vanishes completely.
uses
Finset.sum_add_distrib,Finset.sum_const,nsmul_eq_mul,Finset.mul_sum,Finset.sum_congr,Finset.sum_sub_distrib,mul_zero,add_zerohave hsum_rhs : ∑ i : Fin n, (Real.log (N * (N + 1)) + 2 * (x i - N) / (N + 1)) = (n : ℝ) * Real.log (N * (N + 1)) := by rw [Finset.sum_add_distrib] have hconst : ∑ i : Fin n, Real.log (N * (N + 1)) = (n : ℝ) * Real.log (N * (N + 1)) := by simp [Finset.sum_const, nsmul_eq_mul] rw [hconst] have hzero : ∑ i : Fin n, 2 * (x i - N) / (N + 1) = 0 := by have hfactor : ∑ i : Fin n, 2 * (x i - N) / (N + 1) = (2 / (N + 1)) * ∑ i : Fin n, (x i - N) := by rw [Finset.mul_sum] apply Finset.sum_congr rfl intro i _ ring rw [hfactor] have hsum_sub : ∑ i : Fin n, (x i - N) = 0 := by rw [Finset.sum_sub_distrib] have hxsumN : ∑ i : Fin n, x i = N ^ 2 := by rw [hx_sum] rw [hxsumN] simp [N, Finset.sum_const, nsmul_eq_mul] ring rw [hsum_sub, mul_zero] rw [hzero, add_zero]
ln(∏_{i} (x_i² + N)) = ∑_{i} ln(x_i² + N) and ln((N(N + 1))^n) = n * ln(N(N + 1)).
The logarithm turns the product of positive terms into a sum of logarithms, and pulls the exponent n out of the power (N(N + 1))^n.
uses
Real.log_prod,Real.log_powhave hlogprod_eq : Real.log (∏ i : Fin n, (x i ^ 2 + N)) = ∑ i : Fin n, Real.log (x i ^ 2 + N) := by rw [Real.log_prod] intro i _ exact (hpos i).ne' have htarget_log : Real.log ((N * (N + 1)) ^ n) = (n : ℝ) * Real.log (N * (N + 1)) := by rw [Real.log_pow]
ln(∏_{i} (x_i² + N)) ≤ ln((N(N + 1))^n).
Combining the product-to-sum identity, the evaluated right-hand side, and the sum inequality directly gives the logarithm bound.
have hfinal_log : Real.log (∏ i : Fin n, (x i ^ 2 + N)) ≤ Real.log ((N * (N + 1)) ^ n) := by rw [hlogprod_eq, htarget_log, ← hsum_rhs] exact hsum_le
∏_{i} (x_i² + n) ≤ (n(n + 1))^n.
Both the product ∏ (x_i² + N) and the power (N(N + 1))^n are strictly positive. Since the natural logarithm is strictly increasing on positive reals, ln A ≤ ln B implies A ≤ B. Substituting N = n concludes the proof.
uses
Finset.prod_pos,pow_pos,mul_pos,Real.log_le_log_iffhave hprod_pos : 0 < ∏ i : Fin n, (x i ^ 2 + N) := Finset.prod_pos (fun i _ => hpos i) have hpow_pos : 0 < (N * (N + 1)) ^ n := pow_pos (mul_pos hN0 hN1) n have hle := (Real.log_le_log_iff hprod_pos hpow_pos).mp hfinal_log simpa [N] using hle
Continuous.log- The logarithm of a continuous nowhere-zero real-valued function is continuous.
Continuous.sub- The difference of two continuous functions is continuous.
sq_nonneg- The square of any real number is non-negative.
hasDerivAt_pow- The derivative of x ↦ x^n at x is n * x^(n - 1).
hasDerivAt_add_const_iff- Adding a constant to a function does not change its derivative.
hasDerivAt_const_mul- The derivative of x ↦ c * x at any point is c.
div_eq_mul_inv- Division a / b is equivalent to multiplication by the inverse a * b⁻¹.
mul_comm- Multiplication is commutative: a * b = b * a.
mul_left_comm- Left-commutativity of multiplication: a * (b * c) = b * (a * c).
mul_assoc- Multiplication is associative: (a * b) * c = a * (b * c).
monotoneOn_of_hasDerivWithinAt_nonneg- A continuous function on a convex set whose derivative on the interior is non-negative is monotone increasing on that set.
convex_Icc- A closed interval [a, b] in ℝ is convex.
interior_Icc- The topological interior of the closed interval [a, b] is the open interval (a, b).
sub_nonneg- 0 ≤ a - b if and only if b ≤ a.
div_le_div_iff₀- For positive denominators b, d > 0, a / b ≤ c / d if and only if a * d ≤ c * b.
mul_nonneg- The product of two non-negative numbers is non-negative.
antitoneOn_of_hasDerivWithinAt_nonpos- A continuous function on a convex set whose derivative on the interior is non-positive is antitone (decreasing) on that set.
convex_Ici- A half-infinite interval [a, ∞) in ℝ is convex.
interior_Ici- The topological interior of [a, ∞) is (a, ∞).
sub_nonpos- a - b ≤ 0 if and only if a ≤ b.
le_rfl- Every element in a preorder satisfies a ≤ a.
tournament_degree_prod_le_regular_aux- For m ≥ 2 and x ≥ 1, ln(x² + m) ≤ ln(m(m + 1)) + 2(x - m)/(m + 1).
Finset.sum_le_sum- If f(i) ≤ g(i) for all i in a finite set, then ∑ f(i) ≤ ∑ g(i).
Finset.sum_add_distrib- The sum of a sum of functions distributes over a finite set: ∑ (f(i) + g(i)) = ∑ f(i) + ∑ g(i).
Finset.sum_const- The sum of a constant c over a finite set s equals |s| * c.
nsmul_eq_mul- Scalar multiplication of a real number by a natural number n • a equals (n : ℝ) * a.
Finset.mul_sum- A constant factor can be pulled out of a finite sum: a * ∑ f(i) = ∑ (a * f(i)).
Finset.sum_congr- If two functions agree on every element of a finite set, their sums over that set are equal.
Finset.sum_sub_distrib- The sum of a difference of functions distributes: ∑ (f(i) - g(i)) = ∑ f(i) - ∑ g(i).
mul_zero- Multiplying any element by zero gives zero: a * 0 = 0.
add_zero- Adding zero to any element leaves it unchanged: a + 0 = a.
Real.log_prod- The logarithm of a finite product of non-zero terms is the sum of their logarithms.
Real.log_pow- For real x and natural n, ln(x^n) = n * ln(x).
Finset.prod_pos- A finite product of strictly positive elements is strictly positive.
pow_pos- A strictly positive real number raised to any natural power is strictly positive.
mul_pos- The product of two strictly positive real numbers is strictly positive.
Real.log_le_log_iff- For strictly positive x, y > 0, ln x ≤ ln y if and only if x ≤ y.
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.
lemma tournament_degree_prod_le_regular_aux (m : ℝ) (hm : 2 ≤ m) (x : ℝ) (hx : 1 ≤ x) : Real.log (x ^ 2 + m) ≤ Real.log (m * (m + 1)) + 2 * (x - m) / (m + 1) := by have hm1 : 0 < m + 1 := by linarith let φ : ℝ → ℝ := fun y => Real.log (y ^ 2 + m) - 2 * y / (m + 1) have hcont_log : Continuous (fun y : ℝ => Real.log (y ^ 2 + m)) := by apply Continuous.log · fun_prop · intro y have : 0 < y ^ 2 + m := by nlinarith [sq_nonneg y] exact this.ne' have hcont : Continuous φ := by apply Continuous.sub hcont_log fun_prop have hderiv : ∀ y : ℝ, HasDerivAt φ (2 * y / (y ^ 2 + m) - 2 / (m + 1)) y := by intro y have hy : y ^ 2 + m ≠ 0 := by have : 0 < y ^ 2 + m := by nlinarith [sq_nonneg y] exact this.ne' have h1 : HasDerivAt (fun z : ℝ => z ^ 2 + m) (2 * y) y := by have hbase : HasDerivAt (fun z : ℝ => z ^ 2) (2 * y) y := by simpa using hasDerivAt_pow 2 y exact (hasDerivAt_add_const_iff m).mpr hbase have h2 : HasDerivAt (fun z : ℝ => Real.log (z ^ 2 + m)) (2 * y / (y ^ 2 + m)) y := by simpa using h1.log hy have h3 : HasDerivAt (fun z : ℝ => 2 * z / (m + 1)) (2 / (m + 1)) y := by have h := hasDerivAt_const_mul (x := y) (c := 2 / (m + 1)) simpa [div_eq_mul_inv, mul_comm, mul_left_comm, mul_assoc] using h have hsub := h2.sub h3 change HasDerivAt (fun z : ℝ => Real.log (z ^ 2 + m) - 2 * z / (m + 1)) (2 * y / (y ^ 2 + m) - 2 / (m + 1)) y at hsub simpa [φ] using hsub have hmono : MonotoneOn φ (Set.Icc 1 m) := by apply monotoneOn_of_hasDerivWithinAt_nonneg (convex_Icc (1:ℝ) m) hcont.continuousOn · intro y hy exact (hderiv y).hasDerivWithinAt · intro y hy have hy' : y ∈ Set.Ioo (1:ℝ) m := by simpa [interior_Icc] using hy have hy1 : 1 < y := hy'.1 have hym : y < m := hy'.2 have hden : 0 < y ^ 2 + m := by nlinarith [sq_nonneg y] have hden2 : 0 < m + 1 := hm1 rw [sub_nonneg] exact (div_le_div_iff₀ hden2 hden).mpr (by nlinarith [mul_nonneg (sub_nonneg.mpr hy1.le) (sub_nonneg.mpr hym.le)]) have hanti : AntitoneOn φ (Set.Ici m) := by apply antitoneOn_of_hasDerivWithinAt_nonpos (convex_Ici m) hcont.continuousOn · intro y hy exact (hderiv y).hasDerivWithinAt · intro y hy have hy' : y ∈ Set.Ioi m := by simpa [interior_Ici] using hy have hym : m < y := hy' have hden : 0 < y ^ 2 + m := by nlinarith [sq_nonneg y] have hden2 : 0 < m + 1 := hm1 rw [sub_nonpos] exact (div_le_div_iff₀ hden hden2).mpr (by nlinarith [mul_nonneg (sub_nonneg.mpr (by linarith : 1 ≤ y)) (sub_nonneg.mpr hym.le)]) by_cases hxle : x ≤ m · have hxI : x ∈ Set.Icc (1:ℝ) m := Set.mem_Icc.mpr ⟨hx, hxle⟩ have hmI : m ∈ Set.Icc (1:ℝ) m := Set.mem_Icc.mpr ⟨by linarith, le_rfl⟩ have h := hmono hxI hmI hxle simp only [φ] at h have hφm : m ^ 2 + m = m * (m + 1) := by ring rw [hφm] at h have hsub : 2 * (x - m) / (m + 1) = 2 * x / (m + 1) - 2 * m / (m + 1) := by ring rw [hsub] linarith · have hxge : m ≤ x := by linarith have hxI : x ∈ Set.Ici m := Set.mem_Ici.mpr hxge have hmI : m ∈ Set.Ici m := Set.mem_Ici.mpr le_rfl have h := hanti hmI hxI hxge simp only [φ] at h have hφm : m ^ 2 + m = m * (m + 1) := by ring rw [hφm] at h have hsub : 2 * (x - m) / (m + 1) = 2 * x / (m + 1) - 2 * m / (m + 1) := by ring rw [hsub] linarith/-- Degree product bound for tournament arborescences from concavity. -/theorem tournament_degree_prod_le_regular (n : ℕ) (hn : 2 ≤ n) (x : Fin n → ℝ) (hx_pos : ∀ i, 1 ≤ x i) (hx_sum : ∑ i : Fin n, x i = (n : ℝ) ^ 2) : ∏ i : Fin n, (x i ^ 2 + (n : ℝ)) ≤ ((n : ℝ) * ((n : ℝ) + 1)) ^ n := by let N : ℝ := (n : ℝ) have hN2 : 2 ≤ N := by dsimp [N] exact_mod_cast hn have hN0 : 0 < N := by linarith have hN1 : 0 < N + 1 := by linarith have hpos : ∀ i : Fin n, 0 < x i ^ 2 + N := by intro i have hxi : 1 ≤ x i := hx_pos i nlinarith [sq_nonneg (x i)] have hpoint : ∀ i : Fin n, Real.log (x i ^ 2 + N) ≤ Real.log (N * (N + 1)) + 2 * (x i - N) / (N + 1) := by intro i exact tournament_degree_prod_le_regular_aux N hN2 (x i) (hx_pos i) have hsum_le : ∑ i : Fin n, Real.log (x i ^ 2 + N) ≤ ∑ i : Fin n, (Real.log (N * (N + 1)) + 2 * (x i - N) / (N + 1)) := by exact Finset.sum_le_sum (fun i _ => hpoint i) have hsum_rhs : ∑ i : Fin n, (Real.log (N * (N + 1)) + 2 * (x i - N) / (N + 1)) = (n : ℝ) * Real.log (N * (N + 1)) := by rw [Finset.sum_add_distrib] have hconst : ∑ i : Fin n, Real.log (N * (N + 1)) = (n : ℝ) * Real.log (N * (N + 1)) := by simp [Finset.sum_const, nsmul_eq_mul] rw [hconst] have hzero : ∑ i : Fin n, 2 * (x i - N) / (N + 1) = 0 := by have hfactor : ∑ i : Fin n, 2 * (x i - N) / (N + 1) = (2 / (N + 1)) * ∑ i : Fin n, (x i - N) := by rw [Finset.mul_sum] apply Finset.sum_congr rfl intro i _ ring rw [hfactor] have hsum_sub : ∑ i : Fin n, (x i - N) = 0 := by rw [Finset.sum_sub_distrib] have hxsumN : ∑ i : Fin n, x i = N ^ 2 := by rw [hx_sum] rw [hxsumN] simp [N, Finset.sum_const, nsmul_eq_mul] ring rw [hsum_sub, mul_zero] rw [hzero, add_zero] have hlogprod_eq : Real.log (∏ i : Fin n, (x i ^ 2 + N)) = ∑ i : Fin n, Real.log (x i ^ 2 + N) := by rw [Real.log_prod] intro i _ exact (hpos i).ne' have htarget_log : Real.log ((N * (N + 1)) ^ n) = (n : ℝ) * Real.log (N * (N + 1)) := by rw [Real.log_pow] have hfinal_log : Real.log (∏ i : Fin n, (x i ^ 2 + N)) ≤ Real.log ((N * (N + 1)) ^ n) := by rw [hlogprod_eq, htarget_log, ← hsum_rhs] exact hsum_le have hprod_pos : 0 < ∏ i : Fin n, (x i ^ 2 + N) := Finset.prod_pos (fun i _ => hpos i) have hpow_pos : 0 < (N * (N + 1)) ^ n := pow_pos (mul_pos hN0 hN1) n have hle := (Real.log_le_log_iff hprod_pos hpow_pos).mp hfinal_log simpa [N] using hle