OFFSET
1,4
COMMENTS
The Laurent series for the error of Stirling's approximation of log(n!) (A046968/A046969) is given by the Euler-Maclaurin formula.
Refining Stirling's approximation, by including the first terms in these expansions, makes its logs' errors shrink like 1/n^3, so the sum of log-errors (i.e. log of product of errors) becomes convergent.
log-Gamma's Euler-Maclaurin series comes from applying Watson's lemma to Binet's first log-Gamma formula as a Laplace transform, so it is a rare case in which the converse of Watson's lemma holds; the closed form for the exact value can be found by summing the Laplace transforms and treating the resulting integral as the limit of a Mellin transform.
This is an approximation about n = 0; constants from that about n = -1/2, n = -1 are A389381, A389382. (The formula for Laurent expansions of log(n!) about n = -c was found in p.206 of the Hermite reference.)
The former has < 1/3rd the error, the latter has < 1/5th.
About 0 and -1, the first-order refinement turns a lower bound into an upper one, while about -1/2 it turns an upper into a lower one. These are special cases; the inequality's direction is not always flipped, and about some values the refinement is not an inequality at all (the error's sign is dependent on n).
LINKS
Charles Hermite, Sur la fonction log Γ(a). J. Reine Angew. Math. 115, (1895): 201-208.
Gergő Nemes, Generalization of Binet's Gamma function formulas (2013). Integral Transforms and Special Functions 24, no. 8, 597-606.
Natalia L. Skirrow, Stirling's approximations
FORMULA
EXAMPLE
1.00281160470777211547901529908336...
MAPLE
exp((1 + gamma)/12 - 2*Zeta(1, -1)) / (2*Pi)^(1/4): evalf(%, 100); # Peter Luschny, Oct 02 2025
MATHEMATICA
N[E^((1+EulerGamma)/12-2*Zeta'[-1])/(2*Pi)^(1/4), 128]
CROSSREFS
KEYWORD
nonn,cons
AUTHOR
Natalia L. Skirrow, Sep 28 2025
STATUS
approved
