Faà di Bruno's formula and nested sums of reciprocal squares

TheoremFaà di Bruno's formula for Taylor coefficients#●ready to prove

Let \(f, g : \mathbb{R} \to \mathbb{R}\), let \(a \in \mathbb{R}\) and let \(n\) be a natural number. Suppose that \(g\) is \(n\) times continuously differentiable on a neighbourhood of \(a\) and that \(f\) is \(n\) times continuously differentiable on a neighbourhood of \(g(a)\). For a partition \(\pi\) of \(n\), let \(m_i(\pi)\) be the number of parts of \(\pi\) equal to \(i\), and let \(\ell(\pi) = m_1(\pi) + \cdots + m_n(\pi)\) be its number of parts. Then

\[\frac{(f \circ g)^{(n)}(a)}{n!} = \sum_{\pi} \frac{\ell(\pi)!}{\prod_{i=1}^{n} m_i(\pi)!} \cdot \frac{f^{(\ell(\pi))}(g(a))}{\ell(\pi)!} \cdot \prod_{i=1}^{n} \Bigl(\frac{g^{(i)}(a)}{i!}\Bigr)^{m_i(\pi)},\]

where the sum runs over all partitions \(\pi\) of \(n\).

public theorem_wanted taylorCoeff_comp_eq_sum_partitions (f g : ℝ → ℝ) (a : ℝ) (n : ℕ)
    (hf : ContDiffAt ℝ n f (g a)) (hg : ContDiffAt ℝ n g a) :
    iteratedDeriv n (f ∘ g) a / n.factorial =
      ∑ π : Nat.Partition n,
        ((π.parts.card.factorial : ℝ) /
            ∏ i ∈ Finset.Icc 1 n, ((π.parts.count i).factorial : ℝ)) *
          (iteratedDeriv π.parts.card f (g a) / π.parts.card.factorial) *
          ∏ i ∈ Finset.Icc 1 n, (iteratedDeriv i g a / i.factorial) ^ π.parts.count i
Sources
  • Serge Iovleff, About Some Nested Sums of Reciprocal Squared Numbers, INTEGERS 26 (2026), Article A114, Corollary 1 (quoted as Corollary 4 from the paper's reference [8]).

TheoremNested sums of reciprocal odd squares#●ready to prove

For every natural number \(k\),

\[\sum_{i_1 \ge i_2 \ge \cdots \ge i_k \ge 1} \prod_{j=1}^{k} \frac{1}{(2 i_j - 1)^2} = \frac{(\pi/2)^{2k} E^*_{2k}}{(2k)!},\]

where the sum runs over all weakly decreasing \(k\)-tuples \((i_1, \ldots, i_k)\) of positive integers and \(E^*_{2k} = \sec^{(2k)}(0)\) is the \(2k\)-th derivative at \(0\) of \(\sec x = 1/\cos x\) (the Euler zig number, OEIS A000364). For \(k = 0\) the sum has the single empty term and both sides equal \(1\).

The source also claims that the sum equals \((\pi/4)\,\beta(2k+1)\), where \(\beta\) is Dirichlet's beta function; this is false as printed and is not stated here: at \(k = 0\) the sum is \(1\) but \((\pi/4)\,\beta(1) = \pi^2/16\), and at \(k = 1\) it is \(\pi^2/8\) but \((\pi/4)\,\beta(3) = \pi^4/128\) (the correct factor is \(4/\pi\)).

public theorem_wanted tsum_weaklyDecreasing_prod_inv_odd_sq_eq (k : ℕ) :
    ∑' m : MultipleZeta.WeaklyDecreasingTuple k,
        ∏ j, ((2 * ((m.1 j : ℕ) : ℝ) - 1) ^ 2)⁻¹ =
      (Real.pi / 2) ^ (2 * k) * iteratedDeriv (2 * k) (fun x => (Real.cos x)⁻¹) 0 /
        (2 * k).factorial
Sources
  • Serge Iovleff, About Some Nested Sums of Reciprocal Squared Numbers, INTEGERS 26 (2026), Article A114, Theorem 1.
Uses 1

TheoremNested sums of reciprocal squares#●ready to prove

For every natural number \(k\), the multiple zeta-star value

\[\zeta^\star(\{2\}^k) = \sum_{n_1 \ge n_2 \ge \cdots \ge n_k \ge 1} \frac{1}{n_1^2 n_2^2 \cdots n_k^2},\]

which is \(1\) for \(k = 0\), satisfies

\[\zeta^\star(\{2\}^k) = \frac{2\pi^{2k} |(2^{2k-1} - 1) B_{2k}|}{(2k)!} = 2\,(1 - 2^{1-2k})\,\zeta(2k),\]

where \(B_{2k}\) is the \(2k\)-th Bernoulli number and \(\zeta\) is the Riemann zeta function. The right-hand side is \(2\eta(2k)\) for Dirichlet's eta function \(\eta(s) = (1 - 2^{1-s})\,\zeta(s)\); at \(k = 0\) all three expressions equal \(1\), since \(\zeta(0) = -1/2\).

public theorem_wanted starValue_replicate_two_eq (k : ℕ) :
    MultipleZeta.starValue (MultipleZeta.Index.replicate k 2)
        (MultipleZeta.Index.replicate_isAdmissibleOrEmpty k le_rfl) =
        2 * Real.pi ^ (2 * k) * |((2 : ℝ) ^ ((2 * k : ℤ) - 1) - 1) * bernoulli (2 * k)| /
          (2 * k).factorial ∧
      ((MultipleZeta.starValue (MultipleZeta.Index.replicate k 2)
          (MultipleZeta.Index.replicate_isAdmissibleOrEmpty k le_rfl) : ℝ) : ℂ) =
        2 * ((1 - (2 : ℂ) ^ ((1 : ℤ) - 2 * k)) * riemannZeta (2 * k))
Sources
  • Serge Iovleff, About Some Nested Sums of Reciprocal Squared Numbers, INTEGERS 26 (2026), Article A114, Theorem 2.
Uses 4

TheoremStrictly nested sums of reciprocal odd squares#●ready to prove

For every natural number \(k\),

\[\sum_{i_1 > i_2 > \cdots > i_k \ge 1} \prod_{j=1}^{k} \frac{1}{(2 i_j - 1)^2} = \frac{(\pi/2)^{2k}}{(2k)!},\]

where the sum runs over all strictly decreasing \(k\)-tuples \((i_1, \ldots, i_k)\) of positive integers. For \(k = 0\) the sum has the single empty term and both sides equal \(1\).

The source states the identity for \(k > 0\); the case \(k = 0\) is included here.

public theorem_wanted tsum_strictDecreasing_prod_inv_odd_sq_eq (k : ℕ) :
    ∑' m : MultipleZeta.StrictDecreasingTuple k,
        ∏ j, ((2 * ((m.1 j : ℕ) : ℝ) - 1) ^ 2)⁻¹ =
      (Real.pi / 2) ^ (2 * k) / (2 * k).factorial
Sources
  • Serge Iovleff, About Some Nested Sums of Reciprocal Squared Numbers, INTEGERS 26 (2026), Article A114, Corollary 2.
Uses 1
4 ready to prove

The dependency explorer needs JavaScript and its graph data.

Download the complete graph data.