Asymptotic expansions of power sums
TheoremRamanujan's asymptotic expansion of power sums#✓fully proved
Let \(r \in \mathbb{C}\) with \(r \ne -1\), and let \(N\) be a positive integer with \(\mathrm{Re}\, r < 2N\). For natural \(n\) put \(A_N(n) = \zeta(-r) + \frac{n^{r+1}}{r+1} + \frac{n^r}{2} + \sum_{k=1}^{N} T_k(n)\), where \(T_k(n) = \frac{B_{2k}}{(2k)!}\, r(r-1)\cdots(r-2k+2)\, n^{r-2k+1}\) (a product of \(2k - 1\) factors), \(\zeta\) is the Riemann zeta function, \(B_{2k}\) are the Bernoulli numbers and complex powers are principal. Then \(\sum_{k=1}^{n} k^r - A_N(n) = O(n^{\mathrm{Re}\, r - 2N})\) as \(n \to \infty\).
Proof: Euler--Maclaurin gives the expansion with an integrable periodic-kernel tail. The constant is identified with the Riemann zeta function first in its convergent half-plane, then everywhere needed by analytic continuation.
theorem ramanujan_part1_ch7_entry1_powersum
(r : ℂ) (N : ℕ) (hr : r ≠ -1) (hN : 0 < N) (hu : r.re < 2 * N) :
IsBigO atTop
(fun n : ℕ => chapter7PowerSum r n - chapter7Entry1Approx r N n)
(fun n : ℕ => Real.rpow (n : ℝ) (r.re - 2 * (N : ℝ)))
def chapter7PowerSum (r : ℂ) (n : ℕ) : ℂ :=
∑ k ∈ Icc 1 n, chapter7NatCpow k r
MathlibExt/Analysis/Ramanujan/Part1Ch7Entry1Powersum.lean:74
def chapter7Entry1Approx (r : ℂ) (N n : ℕ) : ℂ :=
riemannZeta (-r) + chapter7NatCpow n (r + 1) / (r + 1) +
chapter7NatCpow n r / 2 + ∑ j ∈ range N, chapter7Entry1Term r n j
MathlibExt/Analysis/Ramanujan/Part1Ch7Entry1Powersum.lean:70
Sources
- Bruce C. Berndt, Ramanujan's Notebooks, Part I, Chapter 7, Entry 1.
Uses 1 · Used by 1
TheoremRamanujan's asymptotic expansion of a binomial combination of log-power sums#✓fully proved
For natural numbers \(m\) and \(x\) let \(S_m(x) = \sum_{k=1}^{x} (\log k)^m\), let \(c_m = (-1)^m \zeta^{(m)}(0)\), where \(\zeta^{(m)}(0)\) is the \(m\)-th derivative at \(0\) of the real function \(s \mapsto \operatorname{Re} \zeta(s)\), and let \(\psi_m(x) = S_m(x) - c_m\). Let \(n\) be a natural number. Then for every \(m \le n\),
as \(x \to \infty\) through the natural numbers, and
as \(x \to \infty\) through the natural numbers, where \(B_j\) are the Bernoulli numbers with \(B_1 = -1/2\).
Proof: Analytic continuation of ζ to re s > -1 by Euler–Maclaurin, Cauchy estimates for the
remainder's derivatives, and the Euler–Maclaurin tail expansion of the log-power sums.
theorem ramanujan_part1_ch8_entry19_gaussdigamma (n : ℕ) :
(∀ m ≤ n,
Tendsto (chapter8LogPowerConstantApprox m) atTop
(𝓝 (chapter8LogPowerConstant m))) ∧
IsBigO atTop
(fun x : ℕ =>
chapter8Entry19Combination n x - chapter8Entry19Approx n x)
(fun x : ℕ => 1 / (x : ℝ) ^ (n + 3))
def chapter8LogPowerConstantApprox (m x : ℕ) : ℝ :=
chapter8LogPowerSum m x -
(∫ t in (1 : ℝ)..(x : ℝ), Real.log t ^ m) -
(1 / 2 : ℝ) * Real.log (x : ℝ) ^ m +
(-1 : ℝ) ^ (m + 1) * (m.factorial : ℝ)
MathlibExt/Analysis/Ramanujan/Part1Ch8Entry19Gaussdigamma.lean:97
def chapter8LogPowerConstant (m : ℕ) : ℝ :=
(-1 : ℝ) ^ m *
iteratedDeriv m (fun s : ℝ => (riemannZeta (s : ℂ)).re) 0
MathlibExt/Analysis/Ramanujan/Part1Ch8Entry19Gaussdigamma.lean:85
def chapter8Entry19Combination (n x : ℕ) : ℝ :=
∑ k ∈ range (n + 1),
(-1 : ℝ) ^ (n + k) * (n.choose k : ℝ) *
Real.log (x : ℝ) ^ k * chapter8Psi (n - k) x
MathlibExt/Analysis/Ramanujan/Part1Ch8Entry19Gaussdigamma.lean:92
def chapter8Entry19Approx (n x : ℕ) : ℝ :=
(n.factorial : ℝ) * x -
(((bernoulli (n + 1) : ℚ) : ℝ) /
((n + 1 : ℕ) * (x : ℝ) ^ n)) -
((n : ℝ) * ((bernoulli (n + 2) : ℚ) : ℝ) /
(2 * (n + 2 : ℕ) * (x : ℝ) ^ (n + 1))) -
((n : ℝ) * ((n : ℝ) + 5 / 3) *
((bernoulli (n + 3) : ℚ) : ℝ) /
(8 * (n + 3 : ℕ) * (x : ℝ) ^ (n + 2)))
MathlibExt/Analysis/Ramanujan/Part1Ch8Entry19Gaussdigamma.lean:72
Sources
- Bruce C. Berndt, Ramanujan's Notebooks, Part I, Chapter 8.
Uses 2
TheoremAsymptotic expansion of partial sums of powers times logarithms#●ready to prove
Let \(r > -1\) be a real number. For a natural number \(n\) write \(\varphi_r(n) = \sum_{k=1}^{n} k^r \log k\) and \(S_r(n) = \sum_{k=1}^{n} k^r\), and for a natural number \(m\) let \(D_m(r) = \sum_{j=0}^{m} \prod_{0 \le q \le m,\ q \ne j} (r - q)\), the derivative of \(r(r-1)\cdots(r-m)\) with respect to \(r\). Let \(B_j\) be the Bernoulli numbers and \(\zeta\) the Riemann zeta function. Then there is a real constant \(C\) such that for every natural number \(N\), as \(n \to \infty\) through the natural numbers,
Moreover, if \(r > 0\), then \(\Gamma(r+1) > 0\), the series \(\sum_{n \ge 1} \log n / n^{r+1}\) converges, and
MathlibExt.Analysis.Ramanujan.Part1Ch9.Entry27Clausen.ramanujan_part1_ch9_entry27_clausen · 22 lines
theorem_wanted ramanujan_part1_ch9_entry27_clausen (r : ℝ) (hr : -1 < r) :
∃ C : ℝ,
(∀ N : ℕ, Asymptotics.IsBigO
(Filter.map (fun n : ℕ => (n : ℝ)) atTop)
(fun x : ℝ =>
chapter9PhiR r x - Real.log x *
(chapter9PowerSumR r x - (riemannZeta (-r : ℂ)).re) -
chapter9Entry27Approx r C N x)
(fun x : ℝ => Real.rpow x (r - 2 * (N : ℝ) - 1))) ∧
(0 < r →
0 < Real.Gamma (r + 1) ∧
Summable (chapter9Entry27LogTerm r) ∧
C =
2 * Real.Gamma (r + 1) * (riemannZeta (r + 1 : ℂ)).re /
Real.rpow (2 * Real.pi) (r + 1) *
(Real.sin (Real.pi * r / 2) *
(Real.log (2 * Real.pi) -
deriv Real.Gamma (r + 1) / Real.Gamma (r + 1)) -
Real.pi / 2 * Real.cos (Real.pi * r / 2)) +
2 * Real.Gamma (r + 1) * Real.sin (Real.pi * r / 2) /
Real.rpow (2 * Real.pi) (r + 1) *
∑' j : ℕ, chapter9Entry27LogTerm r j)
def chapter9PhiR (r x : ℝ) : ℝ :=
∑ k ∈ Icc 1 (Nat.floor x),
Real.rpow (k : ℝ) r * Real.log k
WantedExt/Analysis/Ramanujan/Part1Ch9Wanted.lean:70
def chapter9PowerSumR (r x : ℝ) : ℝ :=
∑ k ∈ Icc 1 (Nat.floor x), Real.rpow (k : ℝ) r
WantedExt/Analysis/Ramanujan/Part1Ch9Wanted.lean:74
def chapter9Entry27Approx (r C : ℝ) (N : ℕ) (x : ℝ) : ℝ :=
C - Real.rpow x (r + 1) / (r + 1) ^ 2 +
∑ j ∈ range N,
let k := j + 1
chapter9Entry27Coeff k r * Real.rpow x (r - 2 * (k : ℝ) + 1)
WantedExt/Analysis/Ramanujan/Part1Ch9Wanted.lean:61
def chapter9Entry27LogTerm (r : ℝ) (j : ℕ) : ℝ :=
Real.log (j + 1 : ℝ) / Real.rpow (j + 1 : ℝ) (r + 1)
WantedExt/Analysis/Ramanujan/Part1Ch9Wanted.lean:67
Sources
- Bruce C. Berndt, Ramanujan's Notebooks, Part I, Chapter 9.
Loading dependency explorer…
The dependency explorer needs JavaScript and its graph data.