Dilogarithm
The dilogarithm function, denoted \operatorname{Li}_2(z), is a special function in complex analysis defined by the power series \sum_{n=1}^\infty \frac{z^n}{n^2} for |z| \leq 1, with analytic continuation to the complex plane via the integral representation \operatorname{Li}_2(z) = -\int_0^z \frac{\ln(1-t)}{t} \, dt, where the principal branch has a branch point at z=1 and a cut along [1, \infty).[1] It serves as the case s=2 of the more general polylogarithm \operatorname{Li}_s(z), and is alternatively known as the Spence function due to its early study by William Spence.[2][3] First introduced by Leonhard Euler in 1768 as part of his investigations into infinite series and integrals, the dilogarithm received its name from Jonathan Hill in 1828, following Spence's 1809 essay that provided the first detailed analysis of its integral form L_2(x) = -\int_0^x \frac{\ln(1-t)}{t} \, dt.[1][4] Subsequent developments by Niels Henrik Abel, Nikolai Lobachevsky, Ernst Kummer, and Srinivasa Ramanujan in the 19th and early 20th centuries uncovered its rich structure, including connections to zeta functions and transcendental number theory.[5][3] The dilogarithm satisfies notable functional equations, such as the duplication formula \operatorname{Li}_2(z) + \operatorname{Li}_2(-z) = \frac{1}{2} \operatorname{Li}_2(z^2) and the reflection relation \operatorname{Li}_2(z) + \operatorname{Li}_2(1-z) = \frac{\pi^2}{6} - \ln z \cdot \ln(1-z) for $0 < z < 1, alongside special values like \operatorname{Li}_2(1) = \frac{\pi^2}{6} and \operatorname{Li}_2\left(\frac{1}{2}\right) = \frac{\pi^2}{12} - \frac{1}{2} \ln^2 2.[1][2] These properties underpin its applications in diverse fields, including algebraic K-theory (via the Bloch-Wigner dilogarithm, a real-valued variant), quantum field theory, and the evaluation of Feynman integrals in particle physics.[5] Modern extensions explore its role in motivic cohomology and string theory amplitudes.[3]Definition and Properties
Definition
The dilogarithm function, denoted \mathrm{Li}_2(z), is a special function arising in mathematics as the case s=2 of the polylogarithm \mathrm{Li}_s(z) = \sum_{n=1}^\infty \frac{z^n}{n^s}.[5] It is defined by the power series \mathrm{Li}_2(z) = \sum_{n=1}^\infty \frac{z^n}{n^2} for complex z with |z| \leq 1. This series converges absolutely throughout the closed unit disk, as the terms satisfy \left| \frac{z^n}{n^2} \right| \leq \frac{1}{n^2} and \sum \frac{1}{n^2} < \infty.[6] An equivalent integral representation is \mathrm{Li}_2(z) = -\int_0^z \frac{\ln(1-t)}{t} \, dt, valid initially within the unit disk. This form was introduced by William Spence in 1809 in his work on logarithmic transcendents.[7]Series Representation
The dilogarithm function, as a special case of the polylogarithm, admits a power series expansion given by\mathrm{Li}_2(z) = \sum_{n=1}^\infty \frac{z^n}{n^2}
for complex z in the unit disk |z| < 1.[5] This series arises from term-by-term integration of the series for the logarithm. Specifically, the Taylor expansion -\ln(1 - u) = \sum_{k=1}^\infty \frac{u^k}{k} for |u| < 1 implies
\frac{-\ln(1 - u)}{u} = \sum_{k=1}^\infty \frac{u^{k-1}}{k} = \sum_{n=0}^\infty \frac{u^n}{n+1}.
Integrating term by term from 0 to z yields
\mathrm{Li}_2(z) = -\int_0^z \frac{\ln(1 - u)}{u} \, du = \sum_{n=0}^\infty \frac{1}{n+1} \int_0^z u^n \, du = \sum_{n=0}^\infty \frac{z^{n+1}}{(n+1)^2} = \sum_{m=1}^\infty \frac{z^m}{m^2},
where the substitution m = n+1 is used, and the interchange of sum and integral is justified by uniform convergence on compact subsets of the unit disk.[5][8] The radius of convergence of this series is 1, as determined by the ratio test:
\lim_{n \to \infty} \left| \frac{z^n / n^2}{z^{n+1} / (n+1)^2} \right| = |z| \lim_{n \to \infty} \frac{(n+1)^2}{n^2} = |z|,
so the series converges absolutely for |z| < 1 and diverges for |z| > 1.[5] At the boundary point z = 1, the series evaluates to the Riemann zeta function at 2,
\mathrm{Li}_2(1) = \sum_{n=1}^\infty \frac{1}{n^2} = \zeta(2) = \frac{\pi^2}{6},
which converges by the p-series test with p = 2 > 1.[5][9] Near z = 0, the asymptotic expansion is simply the partial sums of the series, providing a local approximation; for instance,
\mathrm{Li}_2(z) \approx z + \frac{z^2}{4} + \frac{z^3}{9} + \frac{z^4}{16}
captures the leading behavior for small |z|.[5]
Analytic Continuation
The dilogarithm function \mathrm{Li}_2(z), originally defined by its power series convergence within the unit disk, admits an analytic continuation to the entire complex plane \mathbb{C} excluding a branch cut along the positive real axis from 1 to \infty. This principal branch is constructed using the integral representation \mathrm{Li}_2(z) = -\int_0^z \frac{\ln(1-u)}{u} \, du, where \ln denotes the principal branch of the complex logarithm with branch cut along (-\infty, 0] and argument in (-\pi, \pi]. The resulting function is holomorphic in \mathbb{C} \setminus [1, \infty), real-valued for real arguments z \leq 1, and continuous on the unit circle |z| = 1 via \mathrm{Li}_2(e^{i\theta}) for -\pi < \theta < \pi. For regions where |z| > 1, the analytic continuation is achieved through the inversion functional equation \mathrm{Li}_2(z) = -\mathrm{Li}_2\left(\frac{1}{z}\right) - \frac{[\ln(-z)]^2}{2} + \frac{\pi^2}{6}, with \ln(-z) again on its principal branch. A brief outline of the derivation starts from the integral definition and applies integration by parts, substituting u = 1/w to relate the path to $1/z, yielding the logarithmic term from the boundary contribution at infinity and the constant from the known value \mathrm{Li}_2(1) = \pi^2/6. This equation bridges the exterior of the unit disk back to the interior, where the series applies directly. Encircling the branch point at z=1 induces monodromy in the dilogarithm, reflecting its multi-valued nature. Specifically, crossing the branch cut along [1, \infty) from above to below results in a discontinuity, or jump, of $2\pi i \ln z for real z > 1, arising from the $2\pi i shift in \ln(1-u) when the integration path deforms across the cut. The Spence function, an alternative historical designation for the dilogarithm originating from William Spence's 1809 work, shares this structure and resides on a multi-sheeted Riemann surface to resolve the branches at 0 and 1; the abelian cover of the punctured plane provides a framework for its single-valued extension.Functional Equations
Inversion Formula
The inversion formula for the dilogarithm function provides a key relation that connects the values of \mathrm{Li}_2(z) at z and at $1-z, establishing a fundamental symmetry for $0 < z < 1: \mathrm{Li}_2(z) + \mathrm{Li}_2(1 - z) = \frac{\pi^2}{6} - \ln z \cdot \ln(1 - z). This identity holds on the principal branch and reflects the complementary nature of z and $1-z within the unit interval.[5] To derive this formula, consider the function f(z) = \mathrm{Li}_2(z) + \mathrm{Li}_2(1 - z) + \ln z \cdot \ln(1 - z). Differentiating yields f'(z) = -\frac{\ln(1 - z)}{z} + \frac{\ln z}{1 - z} + \frac{\ln(1 - z)}{z} - \frac{\ln z}{1 - z} = 0, using the known derivative \frac{d}{dz} \mathrm{Li}_2(z) = -\frac{\ln(1 - z)}{z}. Thus, f(z) is constant for $0 < z < 1. Evaluating the limit as z \to 0^+ gives f(0) = \mathrm{Li}_2(1) = \zeta(2) = \pi^2/6, confirming the constant and the formula.[5] This relation extends to complex arguments via analytic continuation, yielding the more general inversion formula \mathrm{Li}_2(z) + \mathrm{Li}_2\left(\frac{1}{z}\right) = -\frac{\pi^2}{6} - \frac{1}{2} [\ln(-z)]^2 for z \in \mathbb{C} \setminus [0, \infty) with \arg(-z) \in (-\pi, \pi), where the logarithm is on the principal branch. The proof follows similarly by differentiation of g(z) = \mathrm{Li}_2(z) + \mathrm{Li}_2(1/z) + \frac{1}{2} [\ln(-z)]^2, which vanishes, and determining the constant using known values such as \mathrm{Li}_2(1) = \pi^2/6. Alternatively, substitute the integral representation \mathrm{Li}_2(z) = -\int_0^z \frac{\ln(1 - t)}{t} \, dt and change variables to relate the forms.[5] These formulas underpin reflection principles in the complex plane, enabling the mapping of dilogarithm values across the unit circle and facilitating computations in regions where direct series evaluation is inefficient, thus enhancing the function's utility in broader analytic contexts.[5]Duplication and Multiplication Relations
The duplication formula relates the dilogarithm at a squared argument to its values at the argument and its negative, providing a basic scaling identity valid for |z| < 1: \mathrm{Li}_2(z^2) = 2 \left[ \mathrm{Li}_2(z) + \mathrm{Li}_2(-z) \right]. [5] This relation follows directly from the power series definition \mathrm{Li}_2(w) = \sum_{k=1}^\infty w^k / k^2 by substituting w = z^2 on the left and expanding the right side, where the even-powered terms double and odd-powered terms cancel, yielding the series for \mathrm{Li}_2(z^2). A generalization of this identity is the distribution relation, which expresses the dilogarithm at an n-th power in terms of a sum over n-th roots of unity: \mathrm{Li}_2(z^n) = n \sum_{k=0}^{n-1} \mathrm{Li}_2(\omega^k z), where \omega = e^{2\pi i / n} is a primitive n-th root of unity, holding for |z| < 1.[5] For n=2, with \omega = -1, this reduces precisely to the duplication formula above.[5] The proof proceeds similarly via series expansion: the sum over roots groups terms in the series for \mathrm{Li}_2(z^n) according to residues modulo n, with the factor of n arising from the orthogonality of the roots of unity in the geometric series sum.[5]Landen Identity and Variants
The Landen identity provides a key functional relation for the dilogarithm, connecting its values at negative and transformed arguments while incorporating a logarithmic term for analytic continuation. It states that \mathrm{Li}_2(-z) = -\mathrm{Li}_2\left(\frac{z}{1+z}\right) - \frac{1}{2} [\ln(1+z)]^2 for z in the complex plane with a branch cut along the negative real axis and |z/(1+z)| < 1. This equation facilitates the evaluation and extension of the dilogarithm beyond its principal series domain.[10] A closely related variant is the duplication relation, which links the dilogarithm at z, -z, and z^2 without logarithmic terms within the unit disk: \mathrm{Li}_2(z) + \mathrm{Li}_2(-z) = \frac{1}{2} \mathrm{Li}_2(z^2). This can be verified directly from the power series expansions for |z| < 1 and serves as a building block for more complex transformations, often combined with the Landen identity in proofs of broader functional equations.[11] For arguments involving hyperbolic functions, the dilogarithm exhibits relations to the inverse tangent integral, defined as \mathrm{Ti}_2(z) = \Im [\mathrm{Li}_2(iz)] for real z with |z| \le 1. This connection arises because \mathrm{Li}_2(iz) yields the imaginary part corresponding to arctangent behaviors, while analytic continuation to real multiples of i links to hyperbolic variants like inverse hyperbolic tangent integrals through identities such as \mathrm{Ti}_2(iz) = -\mathrm{Ti}_2(z) + \frac{1}{2} \ln^2 \left( \frac{1+z}{1-z} \right) for appropriate domains. These forms are particularly useful in integral representations involving trigonometric and hyperbolic substitutions.[12] Abel's generalization extends the Landen identity to a five-term functional equation involving products of dilogarithms, forming the foundation for dilogarithm ladders—iterative chains of relations that evaluate the function at algebraic points through successive transformations. The core Abel equation is \mathrm{Li}_2(x) + \mathrm{Li}_2(y) + \mathrm{Li}_2(1-xy) - \mathrm{Li}_2\left( x(1-y) \right) - \mathrm{Li}_2\left( y(1-x) \right) = \ln x \ln(1-y) + \ln y \ln(1-x) + \ln(1-x) \ln(1-y), valid for x, y \in (0,1) with suitable branch choices; it generalizes two-term relations like Landen's by incorporating cross terms that enable ladder constructions for higher-degree algebraic arguments. These ladders have been instrumental in deriving closed forms for dilogarithms in number-theoretic contexts. The historical development of these identities traces back to the 18th century, with Leonhard Euler exploring dilogarithmic relations in his 1768 paper on infinite series and integrals, where he derived early forms involving logarithmic squares and argument shifts. John Landen anticipated some transformations in his 1760 work on residual analysis, laying groundwork for the modern Landen identity.Special Values and Constants
Evaluations at Rational Points
The dilogarithm function at certain rational points admits closed-form expressions involving the Riemann zeta function value ζ(2) = π²/6 and squares of logarithms. These evaluations are fundamental and arise from the series definition or functional equations such as the inversion relation Li₂(z) + Li₂(1 - z) = π²/6 - ln(z) ln(1 - z).[5] The trivial case is Li₂(0) = 0, as the series terminates immediately.[5] At the boundary of the principal branch, Li₂(1) = ζ(2) = π²/6, which follows from the Basel problem solved by Euler.[5] For negative integers, Li₂(-1) = -π²/12, obtained by substituting z = -1 into the series or using the duplication formula Li₂(z) + Li₂(-z) = (1/2) Li₂(z²).[5] A key rational evaluation is Li₂(1/2) = π²/12 - (1/2) ln²(2), derived by applying the inversion formula at z = 1/2, yielding 2 Li₂(1/2) = π²/6 - ln²(2).[5] Although not strictly rational, the value at φ⁻² = (3 - √5)/2, where φ = (1 + √5)/2 is the golden ratio, is notable for its connection to quadratic irrationals related to rational points via functional equations: Li₂(φ⁻²) = π²/15 - ln²(φ). This arises from considerations in the Bloch group and five-term relations.[5] Multiplication theorems, such as the duplication relation Li₂(z²) = 2 [Li₂(z) + Li₂(-z)], allow extension to other fractions. For example, combining the inversion and duplication formulas yields the closed form Li₂(1/4) = π²/18 - (1/2) ln²(2). Similar techniques provide evaluations at points like 1/3 and 2/3, often involving combinations with higher powers such as Li₂(1/9), as in Ramanujan's identity Li₂(1/3) - (1/6) Li₂(1/9) = π²/18 - (1/6) ln²(3).[5] The following table summarizes representative closed-form evaluations at rational points and related algebraic arguments:| Argument z | Li₂(z) |
|---|---|
| 0 | 0 |
| -1 | -π²/12 |
| 1/2 | π²/12 - (1/2) ln²(2) |
| 1 | π²/6 |
| 1/4 | π²/18 - (1/2) ln²(2) |
| φ⁻² = (3 - √5)/2 | π²/15 - ln²(φ) |
Connections to Zeta Values and Logarithms
The dilogarithm function is fundamentally connected to the Riemann zeta function through its evaluation at unity, where \operatorname{Li}_2(1) = \zeta(2) = \pi^2/6.[1] This relation establishes the dilogarithm as a key special function in analytic number theory, linking its series representation \sum_{k=1}^\infty z^k / k^2 at z=1 directly to the Basel problem solution originally due to Euler.[5] Extensions to higher even zeta values \zeta(2k) involve Bernoulli numbers via the formula \zeta(2k) = (-1)^{k+1} B_{2k} (2\pi)^{2k} / (2 (2k)!), but the dilogarithm specifically captures the weight-2 case, serving as the foundational instance where polylogarithmic functions yield closed-form expressions in terms of \pi^2 and rational coefficients.[1] For complex arguments on the unit circle, the dilogarithm relates to the Clausen function of order 2, defined as \operatorname{Cl}_2(\theta) = \operatorname{Im} \operatorname{Li}_2(e^{i\theta}) = \sum_{n=1}^\infty \sin(n\theta)/n^2.[1] More generally, \operatorname{Li}_2(e^{2\pi i \tau}) for \tau in the upper half-plane connects the imaginary part to \operatorname{Cl}_2(2\pi \{\tau\}) and the real part to related logarithmic terms, providing a bridge between dilogarithmic values and modular forms or eta functions in number theory.[5] This decomposition highlights the dilogarithm's role in expressing transcendental quantities with both polylogarithmic and elementary logarithmic components. The dilogarithm acts as a building block for multiple zeta values through shuffle product relations and iterated integral representations. For instance, the multiple zeta value \zeta(2,1) = \sum_{m>n \geq 1} 1/(m^2 n) = \zeta(3), originally established by Euler via summation techniques, arises in shuffle identities involving products of dilogarithms and logarithms, such as \operatorname{Li}_{1,1}(x,y) + \operatorname{Li}_{1,1}(y,x) + \operatorname{Li}_2(xy) = \operatorname{Li}_1(x) \operatorname{Li}_1(y), where limits as arguments approach 1 recover higher-weight zeta relations.[5][13]Applications and Extensions
Role in Physics
The dilogarithm function arises frequently in the evaluation of two-loop and higher-order Feynman integrals contributing to scattering amplitudes in quantum field theories, where it captures the transcendental structure of the results beyond simple logarithms. In quantum electrodynamics (QED), for instance, the two-loop electron self-energy diagram, which enters corrections to scattering processes like electron-positron annihilation, is expressed in terms of dilogarithms alongside harmonic polylogarithms and iterated integrals. These functions emerge from the integration over loop momenta, providing the finite parts after renormalization and handling branch cuts associated with physical thresholds.[14] In higher-order perturbative calculations, the dilogarithm appears in the evaluation of two-loop self-energy diagrams such as the sunset graph, a scalar two-point function with three propagators that contributes to renormalization in gauge theories like QED and quantum chromodynamics (QCD). The sunset integral, relevant for anomalous dimensions and beta functions in the renormalization group, yields terms involving elliptic dilogarithms in its analytic continuation, particularly when internal masses are equal, linking to modular forms and elliptic curves. These terms ensure the consistency of renormalization procedures by accounting for the transcendental content in the counterterms derived from such diagrams.[15] A notable variant is the Bloch-Wigner dilogarithm, defined as D(z) = \Im \Li_2(z) + \arg(1 - z) \ln |z|, which is real-valued, continuous, and free of branch cuts on the complex plane. This function plays a key role in four-dimensional gauge theories, appearing in the finite parts of Wilson loop expectation values and graphical functions that model non-local correlators. In maximally supersymmetric Yang-Mills theory, for example, it parameterizes the volume of certain polytopes dual to scattering amplitudes, providing a branch-cut-free regularization essential for unitarity and infrared consistency.[16]Use in Number Theory
The Rogers dilogarithm, defined by L(z) = \Li_2(z) + \frac{1}{2} \ln(z) \ln(1 - z) for $0 < z < 1, plays a pivotal role in number theory through its connection to the Bloch group and algebraic K-theory. The Bloch group B(\mathbb{C}) is the kernel of a homomorphism from the free abelian group generated by \mathbb{C}^\times \setminus \{0,1\} to \mathbb{Z}, subject to five-term relations, and it models the structure of K_2(\mathbb{C}), the second algebraic K-group of the complex numbers. The function L(z) induces a regulator homomorphism from B(\mathbb{C}) to \mathbb{R}, mapping generators \{z\} to L(z), which allows for the computation of K_2 invariants and reveals deep arithmetic relations via its functional equations. This framework, developed in works by Suslin and others, links the dilogarithm to Milnor K-theory and provides tools for studying torsion in K-groups.[5][17] In algebraic K-theory, the dilogarithm functions as a regulator mapping elements of K_2(F) for number fields F to real regulators that encode arithmetic data. Specifically, evaluations at roots of unity, such as \Li_2(\zeta) where \zeta is a primitive root of unity, contribute to regulators in the Borel map from K_{2n-1}(F) to the reals, aiding in the determination of ranks and structures related to class numbers of cyclotomic fields. Bloch demonstrated how these evaluations yield explicit formulas for regulators, connecting K-theoretic invariants to L-functions and class group orders in arithmetic geometry. For example, in the context of Stark's conjectures, such regulators help verify predictions about units and class numbers in abelian extensions.[18][19] The dilogarithm also connects to Apéry's constant \zeta(3) through limiting expressions that arise in analytic continuations and series expansions. This limit provides a pathway to express \zeta(3) in terms of dilogarithmic asymptotics, highlighting the function's role in evaluating odd zeta values beyond elementary methods.[5] Historically, Faddeev introduced the quantum dilogarithm in the 1990s as a non-commutative analog, which has applications to modular forms via its modular transformations. The Faddeev quantum dilogarithm \Phi_b(z) satisfies pentagon equations and appears in representations of quantum groups, linking to modular forms through state integrals and wall-crossing phenomena. By 2025, perspectives on its non-perturbative aspects emphasize its use in defining quantum modular forms and holomorphic quantum invariants, where exact evaluations at roots of unity yield relations to L-values and arithmetic modular forms, extending classical dilog identities to quantum settings.[20][21]Generalizations to Polylogarithms
The polylogarithm function, denoted \mathrm{Li}_s(z), is defined by the power series \mathrm{Li}_s(z) = \sum_{n=1}^\infty \frac{z^n}{n^s} for |z| < 1 and complex order s, with the dilogarithm corresponding to the specific case s=2.[1] This series converges absolutely for \operatorname{Re}(s) > 0 within the unit disk, and the function admits an analytic continuation to the complex plane as a multivalued function with a branch point at z=1.[1] In statistical mechanics, the polylogarithm of order s=2 relates directly to Bose-Einstein and Fermi-Dirac integrals, which describe the occupation numbers in ideal quantum gases. Specifically, the Bose-Einstein integral g_2(z) is given by \mathrm{Li}_2(z) for the fugacity z = e^{\mu / kT}, while the Fermi-Dirac integral f_2(z) involves \mathrm{Li}_2(-z), providing closed-form expressions for thermodynamic quantities like pressure and energy density in these systems.[22] A natural extension of the polylogarithm leads to multiple polylogarithms, defined as \mathrm{Li}_{s_1, \dots, s_k}(z_1, \dots, z_k) = \sum_{n_1 > n_2 > \dots > n_k \geq 1} \frac{z_1^{n_1} \cdots z_k^{n_k}}{n_1^{s_1} \cdots n_k^{s_k}} for |z_i| < 1 and positive integers s_i \geq 1, where the single-variable dilogarithm emerges as the depth-one case k=1. These functions generalize the structure of the dilogarithm and play a central role in the study of multiple zeta values and motivic cohomology. Modern extensions include quantum dilogarithm variants, such as the q-deformed dilogarithm \mathrm{Li}_2(q; z), which deform the classical series using a parameter q and arise in the representation theory of quantum groups.[23] These q-deformations preserve key functional relations while adapting to quantum algebraic structures, with applications in integrable systems and knot invariants.[23]Numerical Evaluation
Integral Representations
The dilogarithm function possesses several integral representations that facilitate analytical evaluation and the derivation of functional identities. A fundamental representation is given by the single integral \Li_2(z) = -\int_0^z \frac{\ln(1-t)}{t}\, dt for |z|\leq 1, which serves as the basis for analytic continuation to the complex plane via suitable paths avoiding the branch cut along [1,\infty). This form arises directly from integrating the series definition term by term.[3] An alternative expression utilizes a double integral over the unit square: \Li_2(z) = \int_0^1 \int_0^1 \frac{-\ln u \ln v}{1 - z u v}\, du\, dv, valid for |z|\leq 1, where the denominator expands into a geometric series that aligns with the power series of the dilogarithm. This symmetric form proves advantageous for symmetry-based manipulations and connections to multiple zeta values. These representations enable the derivation of dilogarithm identities through techniques such as parameter differentiation; for instance, introducing a parameter in the limits or integrands and differentiating under the integral sign yields relations like the Landen identity.[3]Computational Algorithms
The dilogarithm function \mathrm{Li}_2(z) for complex arguments with |z| < 1 is computed using its defining power series expansion \mathrm{Li}_2(z) = \sum_{k=1}^{\infty} \frac{z^k}{k^2}, which converges absolutely within the unit disk. To improve efficiency near the boundary where convergence slows, the tail of the series after N terms is approximated using the Euler-Maclaurin formula, expressing the remainder as an integral plus corrections involving Bernoulli numbers; this acceleration yields error bounds on the order of O(z^N / N).[24] For |z| > 1, direct series evaluation is inefficient, so the problem is reduced to the unit disk via the inversion formula \mathrm{Li}_2(z) = -\mathrm{Li}_2\left(\frac{1}{z}\right) - \frac{\pi^2}{6} - \frac{1}{2} \ln^2(-z), where the principal branch of the logarithm is used, and further transformations may map the argument to |w| \leq 1/2 for optimal series convergence; adjustments account for the multi-valued nature of the logarithm to ensure consistency with the desired branch.[24][25] Asymptotic expansions are essential near z = 1, where series methods become poorly conditioned. For z \to 1^-, the reflection identity provides \mathrm{Li}_2(z) = \frac{\pi^2}{6} - \ln(1 - z) \ln z - \mathrm{Li}_2(1 - z), and substituting the series for small $1 - z yields \mathrm{Li}_2(z) \approx \frac{\pi^2}{6} + (1 - z) \ln(1 - z) - (1 - z) + O((1 - z)^2 \ln(1 - z)). For z \to 1^+, the inversion formula combined with the above gives a matching expansion for the real part, approximately \frac{\pi^2}{6} - (z - 1) \ln(z - 1) + O((z - 1)^2 \ln(z - 1)), ensuring continuity of the real part across the cut while the imaginary part jumps by $2\pi i \ln z.[24][5] The dilogarithm's branch cut along the real axis from 1 to \infty requires explicit handling in numerical algorithms to select the principal sheet, where the function is real-valued for real z \leq 1 and the value above the cut satisfies \Im \mathrm{Li}_2(z + i0) = -\pi \ln z for z > 1; computations often use the principal value for the real part and add the appropriate imaginary contribution based on the side of the cut, with variants such as those employing Bernoulli polynomial approximations (e.g., adaptations of Spouge's method for series remainders) to maintain accuracy across sheets.[24][25]Software and Implementations
The dilogarithm function is implemented in Mathematica through thePolyLog[2, z] command, which evaluates the polylogarithm at order 2 for complex argument z. This implementation supports arbitrary-precision arithmetic, allowing computations to any specified decimal precision, and handles branch cut discontinuities along the positive real axis from 1 to \infty in the complex plane, ensuring consistent analytic continuation.[26]
For high-precision computations in C and C++, the Arb library, which extends the GNU Multiple Precision Arithmetic Library (GMP), provides robust support for the dilogarithm as part of its polylogarithm functions. Arb's arb_polylog routine computes \mathrm{Li}_2(z) for real and complex z using ball arithmetic for certified error bounds, enabling evaluations at precisions exceeding thousands of bits while managing branch cuts via analytic continuation methods suitable for numerical applications.[27]
In Python, the mpmath library implements the dilogarithm via mpmath.polylog(2, z), supporting arbitrary-precision floating-point arithmetic with the mpf type. This allows evaluations to high precision, up to approximately $10^6 decimal digits or more, constrained primarily by available memory, and includes handling for complex arguments and principal branch selection.[28]
Recent developments in open-source tools include dedicated Julia packages such as PolyLog.jl, which offers efficient implementations of the real and complex dilogarithm for arbitrary-precision needs. As of Julia 1.12 (released in October 2025) and subsequent updates, the ecosystem—encompassing SpecialFunctions.jl for core special functions and GPU-accelerated libraries like CUDA.jl—facilitates batch evaluations of the dilogarithm on GPUs, enabling high-throughput computations for large arrays of arguments as an alternative to proprietary systems like Mathematica.[29][30]