Documentation

Formalizations.Analysis.SpecialFunctions.Dilogarithm.Basic

The real dilogarithm #

The dilogarithm Li₂ x = ∑_{n ≥ 1} xⁿ / n² on [-1, 1], its derivative -log(1 - x) / x, and the classical functional equations obtained by differentiating and evaluating at an endpoint. Mathlib has no polylogarithm.

Main definitions #

Main results #

References #

noncomputable def Dilog.Li₂ (z : ℝ) :

The real dilogarithm Li₂ z = ∑_{n ≥ 1} zⁿ / n² = ∑_{n ≥ 0} z^{n+1} / (n+1)².

Equations
Instances For
    theorem Dilog.summable_one_div_add_one_sq :
    Summable fun (n : ℕ) => 1 / (↑n + 1) ^ 2

    The dominating series ∑ 1/(n+1)² converges (shift of the p = 2 series).

    theorem Dilog.summable_Li₂ {z : ℝ} (hz : |z| ≤ 1) :
    Summable fun (n : ℕ) => z ^ (n + 1) / (↑n + 1) ^ 2

    The defining series of Li₂ is summable on the closed unit interval |z| ≤ 1 (majorised termwise by the convergent ∑ 1/(n+1)²).

    @[simp]
    theorem Dilog.Li₂_zero :
    Li₂ 0 = 0
    theorem Dilog.Li₂_one :
    Li₂ 1 = Real.pi ^ 2 / 6

    Li₂ 1 = π² / 6 — the Basel problem. Reindex ∑_{n≥0} 1/(n+1)² to hasSum_zeta_two (∑_{n : ℕ} 1/n² = π²/6, whose n = 0 term is 0).

    theorem Dilog.hasDerivAt_Li₂ {x : ℝ} (habs : |x| < 1) (hx0 : x ≠ 0) :

    Termwise differentiation of the dilogarithm series gives, on (-1,1) \ {0}, Li₂'(x) = -log(1 - x)/x (note ∑_{n≥1} xⁿ⁻¹ = 1/(1-x) and ∑ xⁿ/n = -log(1-x)). At x = 0 the derivative is 1, but the quotient form requires x ≠ 0.

    Li₂ is continuous on [-1, 1] (Weierstrass M-test: the series is dominated by ∑ 1/(n+1)² uniformly on the closed interval).

    theorem Dilog.Li₂_add_Li₂_one_sub {x : ℝ} (hx : x ∈ Set.Ioo 0 1) :
    Li₂ x + Li₂ (1 - x) = Real.pi ^ 2 / 6 - Real.log x * Real.log (1 - x)

    Euler's reflection identity for the dilogarithm on (0,1): Li₂ x + Li₂ (1 - x) = π²/6 - log x · log (1 - x).

    Proof: let F y := Li₂ y + Li₂ (1 - y) + log y · log (1 - y). Using hasDerivAt_Li₂ (and (Li₂ (1-·))' = log y/(1-y) by the chain rule), F' = 0 on (0,1), so F is constant there (constant_of_has_deriv_right_zero on subintervals). Its limit as y → 0⁺ is Li₂ 0 + Li₂ 1 + 0 = π²/6 by continuity of Li₂ on [-1,1] and log y · log(1-y) = (log(1-y)/y)·(y·log y) → (-1)·0 = 0.

    theorem Dilog.Li₂_one_half :
    Li₂ (1 / 2) = Real.pi ^ 2 / 12 - Real.log 2 ^ 2 / 2

    Euler's special value Li₂(1/2) = π²/12 - log²2/2 — the reflection identity at its fixed point x = 1/2.

    theorem Dilog.Li₂_landen {x : ℝ} (hx : x ∈ Set.Ioo 0 (1 / 2)) :
    Li₂ x + Li₂ (x / (x - 1)) = -Real.log (1 - x) ^ 2 / 2

    Landen's identity for the dilogarithm on (0, 1/2): Li₂ x + Li₂ (x/(x-1)) = -log²(1-x) / 2.

    For x ∈ (0, 1/2) the second argument x/(x-1) ∈ (-1, 0) stays inside the disc of convergence of the defining power series; past x = 1/2 the identity requires analytic continuation, which the series definition does not provide.

    Proof: same scheme as the reflection identity — the combination G y := Li₂ y + Li₂ (y/(y-1)) + log²(1-y)/2 has zero derivative on (0, 1/2) and tends to 0 as y → 0⁺.

    theorem Dilog.Li₂_sq {x : ℝ} (hx : |x| ≤ 1) :
    Li₂ (x ^ 2) = 2 * (Li₂ x + Li₂ (-x))

    Duplication (squaring) formula Li₂(x²) = 2(Li₂ x + Li₂(−x)) for |x| ≤ 1 — pure series rearrangement: odd powers cancel, even powers double. No calculus needed.

    Golden-ratio evaluations #

    The classical special values Li₂(1/φ²) = π²/15 − log²φ, Li₂(1/φ) = π²/10 − log²φ and Li₂(−1/φ) = log²φ/2 − π²/15 (attributed to Landen; Lewin §1.4). The key golden facts are 1/φ + 1/φ² = 1 (so 1/φ² and 1/φ are a reflection pair) and (1/φ²)/((1/φ²)−1) = −1/φ (so Landen's identity maps 1/φ² to −1/φ); together with the duplication formula at 1/φ this gives a linear system — no five-term relation is needed.

    theorem Dilog.Li₂_neg_one :
    Li₂ (-1) = -Real.pi ^ 2 / 12

    Li₂(−1) = −π²/12 = −η(2): the duplication formula at x = 1.

    Li₂(1/φ²) = π²/15 − log²φ (classical; Landen, cf. Lewin §1.4).

    Li₂(1/φ) = π²/10 − log²φ (classical; Landen, cf. Lewin §1.4).

    Li₂(−1/φ) = log²φ/2 − π²/15 (classical; Lewin §1.4).

    noncomputable def Dilog.rogersL (x : ℝ) :

    The Rogers dilogarithm L(x) = Li₂(x) + ½·log x · log(1−x) (the natural normalisation for 0 < x < 1; the log-product term removes the log² defects, so the golden-ratio values become pure rational multiples of π²).

    Equations
    Instances For

      L(1/φ²) = π²/15 — in physics terms: the effective central charge of the Lee-Yang thermodynamic-Bethe-ansatz system is c_eff = L(1/φ²)/L(1) = (π²/15)/(π²/6) = 2/5, the simplest TBA dilogarithm identity.

      Rogers' two-term golden identity L(1/φ) + L(1/φ²) = L(1) = π²/6 — the weight-2 "sum rule" behind the golden-ratio ladder (Rogers 1907).

      L(1) = Li₂(1) = π²/6: the log-product term vanishes at x = 1.

      The Lee–Yang effective central charge c_eff = L(1/φ²)/L(1) = 2/5, the simplest thermodynamic-Bethe-ansatz dilogarithm identity.