Watson's cosecant and cosine sums

DefinitionGeneralized Watson trigonometric sum#✓fully proved

For natural numbers \(n\) and \(\nu\), \(C_n(\nu)\) is the real number

\[ C_n(\nu) = \sum_{l=1}^{n-1} \frac{\cos(2\pi\nu l/n)}{\sin(\pi l/n)}. \]

For \(n \le 1\) the sum is empty and \(C_n(\nu) = 0\).

public noncomputable def watsonCosSum (n ν : ℕ) : ℝ :=
  ∑ l ∈ Finset.Icc 1 (n - 1),
    Real.cos (2 * Real.pi * (ν : ℝ) * (l : ℝ) / (n : ℝ)) *
      (1 / Real.sin (Real.pi * (l : ℝ) / (n : ℝ)))
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, definition of C_n(ν).
Used by 3

DefinitionWatson's alternating cosecant sum#✓fully proved

For a natural number \(n\), the alternating cosecant sum is the real number

\[ C_n = \sum_{l=1}^{n-1} \frac{(-1)^{l+1}}{\sin(\pi l / n)}. \]

For \(n \le 1\) the sum is empty and \(C_n = 0\).

public noncomputable def watsonAlternatingCosecantSum (n : ℕ) : ℝ :=
  ∑ l ∈ Finset.Icc 1 (n - 1),
    (-1 : ℝ) ^ (l + 1) * (1 / Real.sin (Real.pi * (l : ℝ) / (n : ℝ)))
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, definition of C_n.
Used by 5

TheoremWatson's cosine sum at \(n = 2\), \(\nu = 1\)#✓fully proved

The cosine sum \(C_n(\nu) = \sum_{l=1}^{n-1} \cos(2 \pi \nu l / n) / \sin(\pi l / n)\) satisfies \(C_2(1) = -1\).

public theorem watsonCosSum_two_one : watsonCosSum 2 1 = -1
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, definition of C_n(ν).
Uses 1

TheoremWatson's alternating cosecant sum at \(n = 2\)#✓fully proved

The alternating cosecant sum \(C_2 = \sum_{l=1}^{1} (-1)^{l+1} / \sin(\pi l / 2)\) equals \(1\).

public theorem watsonAlternatingCosecantSum_two : watsonAlternatingCosecantSum 2 = 1
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, definition of C_n.
Uses 1

DefinitionOdd cotangent derivatives#✓fully proved

For natural numbers \(r\), \(n\) and \(\nu\), \(\mathcal{H}_{2r-1}(\nu/n)\) is the real number \(\cot^{(2r-1)}(\pi\nu/n)\), the \((2r-1)\)st derivative of the cotangent function evaluated at \(\pi\nu/n\). Here \(2r - 1\) is computed in the natural numbers, so for \(r = 0\) this is \(\cot(\pi\nu/n)\) itself.

The derivative is taken of the cotangent as a function on all of \(\mathbb{R}\); the value is the classical one whenever \(\pi\nu/n\) is not a multiple of \(\pi\).

public noncomputable def watsonOddCotangentDeriv (r n ν : ℕ) : ℝ :=
  iteratedDeriv (2 * r - 1) Real.cot (Real.pi * (ν : ℝ) / (n : ℝ))
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, definition of ℋ_{2r-1}.
Used by 2

TheoremDigamma evaluation of a hyperbolic Laplace integral#●ready to prove

Let \(b > 0\) be a real number and let \(\alpha, \beta\) be complex numbers with \(\operatorname{Re}(\alpha + b - \beta) > 0\) and \(\operatorname{Re}(\alpha + b + \beta) > 0\). Then the function \(x \mapsto e^{-\alpha x} \cosh(\beta x) / \cosh(bx)\) is integrable on \((0, \infty)\), and, writing \(\psi\) for the digamma function,

\[ \int_0^\infty \frac{e^{-\alpha x} \cosh(\beta x)}{\cosh(bx)} \, dx = \frac{1}{4b} \Bigl[ \psi\bigl(\tfrac34 + \tfrac{\alpha+\beta}{4b}\bigr) - \psi\bigl(\tfrac14 + \tfrac{\alpha-\beta}{4b}\bigr) + \psi\bigl(\tfrac34 + \tfrac{\alpha-\beta}{4b}\bigr) - \psi\bigl(\tfrac14 + \tfrac{\alpha+\beta}{4b}\bigr) \Bigr]. \]

The source prints only the condition \(\operatorname{Re}(\alpha + b - \beta) > 0\); the condition \(\operatorname{Re}(\alpha + b + \beta) > 0\) is also needed, since the integrand is even in \(\beta\).

MetaMathlibExt.watson_cosh_integral_digamma · 17 lines
public theorem_wanted watson_cosh_integral_digamma
    (b : ℝ) (α β : ℂ) (hb : 0 < b)
    (hminus : 0 < (α + (b : ℂ) - β).re)
    (hplus : 0 < (α + (b : ℂ) + β).re) :
    IntegrableOn
        (fun x : ℝ =>
          Complex.exp (-α * (x : ℂ)) * Complex.cosh (β * (x : ℂ)) /
            Complex.cosh ((b : ℂ) * (x : ℂ)))
        (Set.Ioi 0) ∧
      (∫ (x : ℝ) in Set.Ioi 0,
          Complex.exp (-α * (x : ℂ)) * Complex.cosh (β * (x : ℂ)) /
            Complex.cosh ((b : ℂ) * (x : ℂ))) =
        (1 / (4 * (b : ℂ))) *
          (Complex.digamma ((3 / 4 : ℂ) + (α + β) / (4 * (b : ℂ))) -
            Complex.digamma ((1 / 4 : ℂ) + (α - β) / (4 * (b : ℂ))) +
            Complex.digamma ((3 / 4 : ℂ) + (α - β) / (4 * (b : ℂ))) -
            Complex.digamma ((1 / 4 : ℂ) + (α + β) / (4 * (b : ℂ))))
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, Lemma 1.

DefinitionWatson expansion terms#✓fully proved

For natural numbers \(r\), \(n\) and \(\nu\), let \(B_{2r}\) be the \(2r\)th Bernoulli number and let \(\mathcal{H}_{2r-1}(\nu/n) = \cot^{(2r-1)}(\pi\nu/n)\) be the \((2r-1)\)st derivative of the cotangent evaluated at \(\pi\nu/n\). The \(r\)th Watson expansion term is the real number

\[ \Bigl(1 - \frac{1}{2^{2r-1}}\Bigr) \frac{\pi^{2r-1} B_{2r} \, \mathcal{H}_{2r-1}(\nu/n)}{(2r)! \, n^{2r-1}}. \]

Here \(2r - 1\) is computed in the natural numbers, so for \(r = 0\) the factor \(1 - 1/2^{2r-1}\) is \(0\) and the term vanishes.

public noncomputable def watsonExpansionTerm (r n ν : ℕ) : ℝ :=
  (1 - 1 / (2 : ℝ) ^ (2 * r - 1)) * Real.pi ^ (2 * r - 1) *
      (((bernoulli (2 * r) : ℚ) : ℝ)) * watsonOddCotangentDeriv r n ν /
    ((Nat.factorial (2 * r) : ℝ) * (n : ℝ) ^ (2 * r - 1))
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, Theorem 3 (printed as Theorem 4).
Uses 1 · Used by 1

TheoremAlmost-asymptotic expansion of Watson's cosine sum#●ready to prove

Let \(n \ge 2\), \(N \ge 2\) and \(\nu\) be natural numbers with \(1 \le \nu \le n - 1\), and let

\[ C_n(\nu) = \sum_{l=1}^{n-1} \frac{\cos(2 \pi \nu l / n)}{\sin(\pi l / n)}. \]

For \(r \ge 1\) write

\[ T_r(n, \nu) = \frac{(1 - 2^{1-2r}) \, \pi^{2r-1} B_{2r} \, \cot^{(2r-1)}(\pi \nu / n)} {(2r)! \, n^{2r-1}}, \]

where \(B_{2r}\) is the \(2r\)-th Bernoulli number and \(\cot^{(2r-1)}\) is the \((2r-1)\)-st derivative of the cotangent. Then there is a real number \(\lambda\) with \(0 < \lambda < 1\) such that

\[ C_n(\nu) = -\frac{2n}{\pi} \log\left(2 \sin\frac{\pi \nu}{n}\right) + 2 \sum_{r=1}^{N-1} T_r(n, \nu) + 2 \lambda \, T_N(n, \nu). \]
public theorem_wanted watson_cos_sum_almost_asymptotic
    (n ν N : ℕ) (hn : 2 ≤ n) (hN : 2 ≤ N) (hνpos : 1 ≤ ν)
    (hνmax : ν ≤ n - 1) :
    ∃ remainder : ℝ, 0 < remainder ∧ remainder < 1 ∧
      watsonCosSum n ν =
        -(2 * (n : ℝ) / Real.pi) *
            Real.log (2 * Real.sin (Real.pi * (ν : ℝ) / (n : ℝ))) +
          2 * (∑ r ∈ Finset.Icc 1 (N - 1), watsonExpansionTerm r n ν) +
          2 * remainder * watsonExpansionTerm N n ν
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, Theorem 3 (printed as Theorem 4).
Uses 2

TheoremThe odd cotangent derivatives in Watson's expansion are negative#●ready to prove

Let \(r\), \(n\) and \(\nu\) be natural numbers with \(r \ge 1\) and \(1 \le \nu \le n - 1\). Then \(\mathcal{H}_{2r-1}(\nu/n) = \cot^{(2r-1)}(\pi\nu/n)\), the \((2r-1)\)st derivative of the cotangent evaluated at \(\pi\nu/n\), is strictly negative.

public theorem_wanted watson_odd_cotangent_deriv_neg
    (r n ν : ℕ) (hr : 1 ≤ r) (hνpos : 1 ≤ ν) (hνmax : ν ≤ n - 1) :
    watsonOddCotangentDeriv r n ν < 0
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, Theorem 3 (printed as Theorem 4).
Uses 1

DefinitionMain part of the asymptotic expansion of Watson's alternating cosecant sum#✓fully proved

For natural numbers \(N\) and \(n\), the main part is the real number

\[ M_N(n) = \frac{2 n \log 2}{\pi} + 2 \sum_{r=1}^{N-1} \frac{(-1)^{r+1} (2^{2r-1} - 1)(2^{2r} - 1) \pi^{2r-1} B_{2r}^2}{r \, (2r)! \, n^{2r-1}}, \]

where \(B_{2r}\) is the \(2r\)-th Bernoulli number. For \(N \le 1\) the sum is empty.

public noncomputable def watsonAlternatingMain (N n : ℕ) : ℝ :=
  2 * (n : ℝ) * Real.log 2 / Real.pi +
    2 * (∑ r ∈ Finset.Icc 1 (N - 1),
      (-1 : ℝ) ^ (r + 1) * ((2 : ℝ) ^ (2 * r - 1) - 1) *
          ((2 : ℝ) ^ (2 * r) - 1) * Real.pi ^ (2 * r - 1) *
          (((bernoulli (2 * r) : ℚ) : ℝ) ^ 2) /
        ((r : ℝ) * (Nat.factorial (2 * r) : ℝ) * (n : ℝ) ^ (2 * r - 1)))
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, Corollary 2.
Used by 1

TheoremComplete asymptotic expansion of Watson's alternating cosecant sum#●ready to prove

For natural numbers \(N\) and \(n\) let \(C_n = \sum_{l=1}^{n-1} (-1)^{l+1} / \sin(\pi l / n)\) and

\[ M_N(n) = \frac{2 n \log 2}{\pi} + 2 \sum_{r=1}^{N-1} \frac{(-1)^{r+1} (2^{2r-1} - 1)(2^{2r} - 1) \pi^{2r-1} B_{2r}^2}{r \, (2r)! \, n^{2r-1}}, \]

where \(B_{2r}\) is the \(2r\)-th Bernoulli number. For every \(N \ge 2\) there is a real constant \(K \ge 0\) such that for every even \(n \ge 2\),

\[ |C_n - M_N(n)| \le \frac{K}{n^{2N-1}}. \]

In other words, \(C_n = M_N(n) + O(n^{1-2N})\) as \(n \to \infty\) through even values.

public theorem_wanted watson_alternating_complete_asymptotic (N : ℕ) (hN : 2 ≤ N) :
    ∃ K : ℝ, 0 ≤ K ∧ ∀ n : ℕ, Even n → 2 ≤ n →
      |watsonAlternatingCosecantSum n - watsonAlternatingMain N n| ≤
        K / (n : ℝ) ^ (2 * N - 1)
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, Corollary 2.
Uses 2

TheoremWatson's alternating cosecant sum vanishes for odd \(n\)#●ready to prove

For every odd natural number \(n\),

\[ \sum_{l=1}^{n-1} \frac{(-1)^{l+1}}{\sin(\pi l / n)} = 0. \]
public theorem_wanted watson_alternating_odd_zero (n : ℕ) (hn : Odd n) :
    watsonAlternatingCosecantSum n = 0
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, Corollary 2 and Theorem 6.
Uses 1

TheoremTwo-sided bounds for Watson's alternating cosecant sum#●ready to prove

For every even natural number \(n \ge 2\), the alternating cosecant sum \(C_n = \sum_{l=1}^{n-1} (-1)^{l+1} / \sin(\pi l / n)\) satisfies the strict inequalities

\[ \frac{2 n \log 2}{\pi} + \frac{\pi}{12 n} - \frac{7 \pi^3}{1440 n^3} < C_n < \frac{2 n \log 2}{\pi} + \frac{\pi}{12 n}. \]

The lower bound consists of the first three terms of the asymptotic expansion of \(C_n\) and the upper bound of the first two.

public theorem_wanted watson_alternating_sharp_bounds
    (n : ℕ) (heven : Even n) (hn : 2 ≤ n) :
    2 * (n : ℝ) * Real.log 2 / Real.pi + Real.pi / (12 * (n : ℝ)) -
          7 * Real.pi ^ 3 / (1440 * (n : ℝ) ^ 3) <
        watsonAlternatingCosecantSum n ∧
      watsonAlternatingCosecantSum n <
        2 * (n : ℝ) * Real.log 2 / Real.pi + Real.pi / (12 * (n : ℝ))
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, Theorem 6 and Corollary 3.
Uses 1

TheoremDecimal bounds for Watson's alternating cosecant sum#●ready to prove

For every even natural number \(n \ge 4\), the alternating cosecant sum \(C_n = \sum_{l=1}^{n-1} (-1)^{l+1} / \sin(\pi l / n)\) satisfies the strict inequalities

\[ \frac{2 n \log 2}{\pi} + \frac{0.252}{n} < C_n < \frac{2 n \log 2}{\pi} + \frac{0.262}{n}. \]
public theorem_wanted watson_alternating_simple_bounds
    (n : ℕ) (heven : Even n) (hn : 4 ≤ n) :
    2 * (n : ℝ) * Real.log 2 / Real.pi + (0.252 : ℝ) / (n : ℝ) <
        watsonAlternatingCosecantSum n ∧
      watsonAlternatingCosecantSum n <
        2 * (n : ℝ) * Real.log 2 / Real.pi + (0.262 : ℝ) / (n : ℝ)
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, Corollary 3.
Uses 1

TheoremVanishing limits of Watson's cosine sum at one sixth and five sixths#●ready to prove

For natural numbers \(n\) and \(\nu\) let \(C_n(\nu) = \sum_{l=1}^{n-1} \cos(2 \pi \nu l / n) / \sin(\pi l / n)\). As \(m \to \infty\) through the natural numbers, \(C_{6m}(m) \to 0\) and \(C_{6m}(5m) \to 0\).

public theorem_wanted watson_cos_sum_sixth_limits :
    Tendsto (fun m : ℕ => watsonCosSum (6 * m) m) atTop (𝓝 0) ∧
      Tendsto (fun m : ℕ => watsonCosSum (6 * m) (5 * m)) atTop (𝓝 0)
Sources
  • Iaroslav V. Blagouchine, On a Generalization of Watson’s Trigonometric Sum (On Dowker’s Sum of Order One Half), INTEGERS 25 (2025), Article A30, Corollary 4.
Uses 1
7 fully proved8 ready to prove

The dependency explorer needs JavaScript and its graph data.

Download the complete graph data.