Evaluating a Definite Integral with a ln x Denominator

A cyclotomic-style denominator and a ln x term look like a convergence problem but are not. The real work is tracking one parameter shift and one factor of 1/6 through a digamma-to-log-gamma step.

Binary Buddies · 5 min read · 2 views

The integral ∫₀¹ (x¹³ − 1) / (ln x · (x² + x + 1)(x² − x + 1)) dx looks like it should fail twice. ln x is zero at x = 1, and the quadratic factors look like trouble. Neither is a problem. x¹³ − 1 vanishes at 1 at the same rate as ln x, so the integrand tends to 13/3 there, and both quadratics have only complex roots. The integrand is continuous on the whole interval.

The hard part is the closed form. The original Math StackExchange question gets as far as a digamma series and then stalls on integrating it back through log-gamma values at sixths. The value is ln 5 − (1/3) ln 2 ≈ 1.37839.

This article follows the route that gets there, and says plainly where it still needs special functions.

Collapsing the denominator

The two quadratics multiply to x⁴ + x² + 1. Multiplying top and bottom by x² − 1 turns that into x⁶ − 1, so the integral becomes:

I = ∫₀¹ (x¹³ − 1)(1 − x²) / ((1 − x⁶) ln x) dx

Flipping the sign of both x² − 1 and x⁶ − 1 changes nothing, and 1 − x⁶ is the form that expands cleanly later. A quick sanity check: on (0, 1), x¹³ − 1 and ln x are both negative, so the integrand is positive. The answer therefore has to be positive, and 1.37839 is.

Parameterising the exponent

The accepted answer defines:

f(s) = ∫₀¹ (x^(s−1) − 1)/ln x · (1 − x²)/(1 − x⁶) dx

The target is f(14), since s − 1 = 13. This is an easy place to be off by one: the question's own parameter a equals s − 1, so the upper limit of integration in a is 13 but in s it is 14. Also, f(1) = 0 because the numerator is x⁰ − 1, which gives the initial condition for free.

Differentiating in s brings down a factor of ln x that cancels the denominator:

f'(s) = ∫₀¹ (x^(s−1) − x^(s+1)) / (1 − x⁶) dx

From the series to digamma

Expanding 1/(1 − x⁶) as a geometric series and integrating term by term gives Σ [1/(6n + s) − 1/(6n + s + 2)]. The answer recognises this as a difference of digamma values, using ∫₀¹ (1 − xᵗ)/(1 − x) dx = ψ(t + 1) + γ:

f'(s) = [ψ((s + 2)/6) − ψ(s/6)] / 6

The factor of 1/6 is exactly the chain-rule factor from d/ds ln Γ(s/6) = ψ(s/6)/6, so the antiderivative is a difference of log-gamma terms with no stray constants. Fixing the constant with f(1) = 0 and Γ(1/2) = √π gives:

f(s) = ln [ Γ((s + 2)/6) · Γ(1/6) / ( Γ(s/6) · √π ) ]

This is the step the asker was dreading, and it turns out to be one line once the 1/6 is tracked.

Reducing the gamma values to logarithms

At s = 14 the formula needs Γ(8/3)/Γ(7/3). The recurrence Γ(z + 1) = zΓ(z) reduces it: Γ(8/3) = (5/3)(2/3)Γ(2/3) and Γ(7/3) = (4/3)(1/3)Γ(1/3). The ratio is (5/2) · Γ(2/3)/Γ(1/3), which is where the factor 5/2 in the answer comes from.

The remaining ratio is handled by the duplication and reflection formulas, which the answer quotes as Γ(2/3)/Γ(1/3) = 2^(2/3) √π / Γ(1/6). The Γ(1/6) and √π cancel against the constant factor, leaving:

f(14) = ln( (5/2) · 2^(2/3) ) = ln 5 − (1/3) ln 2

Numerically, ln 5 ≈ 1.60944 and (1/3) ln 2 ≈ 0.23105, which gives 1.37839. That matches the numerical value the answer reports, and it is a useful independent check on the gamma identities.

The series route and its limit

A second answer skips the parameter and expands directly. Every partial sum is nonnegative and increases pointwise, so monotone convergence justifies swapping sum and integral. That is a more careful justification than differentiating under the integral sign, which is usually just asserted. Each term is then evaluated with ∫₀¹ (xᵃ − xᵇ)/ln x dx = ln((a + 1)/(b + 1)):

I = Σ ln( (6n + 14)(6n + 3) / ((6n + 16)(6n + 1)) )

This is the same integral as an infinite product. The product still has to be evaluated, so this route does not by itself produce the closed form. The digamma route above does.

What to watch for

The accepted route does not avoid special functions, which the asker hoped for. It only makes them tidy: the (1 − x²) factor makes f'(s) a difference of just two digamma terms, and the final step is a handful of Gamma identities. A different numerator or denominator could leave more terms, so don't assume the same cancellation.

Three practical checks. First, keep the shift between a and s straight, or you will evaluate f(13) and get a wrong but plausible-looking logarithm. Second, confirm the sign of the integrand before trusting any closed form. Third, always compare against a numerical value, here 1.37839, because an error in a Gamma identity can produce a result that is clean-looking and wrong.

This article draws on Evaluating the definite integral $\int_0^1 \frac{x^{13} - 1}{\ln x (x^2 + x + 1)(x^2 - x + 1)} \, dx$ and answers to it (1, 2), on the Stack Exchange network, licensed under CC BY-SA 4.0.

  • #Feynman's trick
  • #Gamma function
  • #Integration
  • #calculus
  • #definite-integrals
  • #digamma function

Comments

Add a comment
Sign in with Google to join the conversation