Decomposition Formulas for Multiple Polylogarithms

• The paper presents explicit reduction formulas that reduce multiple polylogarithms (up to weight four) to a standard basis including logarithms, Li₂, Li₃, Li₄, and Li₂,₂ using shuffle and symbol calculus methods.
• It employs algebraic tools such as shuffle, stuffle, and coproduct decompositions to uncover functional relations and symmetry-based identities among iterated integrals.
• The findings have significant implications in arithmetic geometry, quantum field theory, and combinatorics by streamlining the evaluation of complex polylogarithmic expressions.

Multiple polylogarithms admit a remarkable array of algebraic and functional decomposition formulas, with deep significance across arithmetic geometry, quantum field theory, and the combinatorics of iterated integrals. Central strands in this field include the explicit reduction of polylogarithmic iterated integrals to algebraic bases, the combinatorics of shuffle/stuffle and symbol tensor structures, functional equations arising from symmetry, and interrelations among classical polylogarithms, multiple zeta values, and generalized functions such as $Li_{2,2}$. These decomposition results structure the functional and algebraic properties of polylogarithms at all weights and unveil hidden algebraic relations critical in contemporary research.

1. Decomposition of Generalized Polylogarithms up to Weight Four

For generalized (Goncharov) polylogarithms (GPLs) $G(a_1,\ldots,a_n;x)$, all functions up to weight four can be reduced to a standard basis involving logarithms, classical polylogarithms, and the double polylogarithm

$$\log(x), \quad Li_2(x), \quad Li_3(x), \quad Li_4(x), \quad Li_{2,2}(x,y)$$

with all branch cuts tracked by piecewise constant triangle functions and suitable sign corrections. The key methodology utilizes shuffle and stuffle products, rescaling transformations, the $1-x$ involution, and integration-by-parts. The reductions are as follows (Frellesvig et al., 2016):

• Weight 1 ($n=1$):

$$G(;x) = 1,\,\;\; G(0;x) = \ln x,\,\;\; G(a;x) = \ln(1-x/a)\quad (a\neq 0)$$

• Weight 2 ($n=2$):

$$\begin{aligned} G(0,0;x) &= \tfrac12 \ln^2 x,\;\; G(0,a;x) = -Li_2(x/a),\ G(a,0;x) &= \ln x \ln(1-x/a) + Li_2(x/a),\ G(a,a;x) &= \tfrac12 \ln^2(1-x/a) \end{aligned}$$ More generally,

$$G(a,b;x) = Li_2\Big(\frac{b-x}{b-a}\Big) - Li_2\Big(\frac{b}{b-a}\Big) + \ln\Big(1 - \frac{x}{b}\Big) \ln\Big(\frac{x - a}{b - a}\Big) + \ldots$$

where omitted terms account for discontinuities.

• Weight 3 ($n=3$):

All GPLs reduce to $\ln$, $Li_2$, $Li_3$. For instance:

$$G(0,0,0;x) = \tfrac16 \ln^3 x,\;\; G(0,0,a;x) = -Li_3(x/a)$$

and more generally, 3-letter functions at argument 1 are reduced using $1-x$ transformations and symbol calculus.

• Weight 4 ($n=4$):

Any weight-4 GPL is a combination of

$$\ln(x),\, Li_2(x),\, Li_3(x),\, Li_4(x),\, Li_{2,2}(x,y)$$

The function $$Li_{2,2}(x, y) = \sum_{m > n > 0} \frac{x^m}{m^2} \frac{y^n}{n^2}$$ is irreducible to single-variable polylogs. Reduction proceeds via shuffle, $1-x$ involution, and symbol techniques, isolating all higher-depth, weight-4 GPLs as explicit combinations of $Li_4$ and $Li_{2,2}$, plus lower-weight/logarithmic terms (Frellesvig et al., 2016).

2. Symbol Calculus and Functional Decomposition at Weight Four

In weight-4, the symbol method provides a functorial framework to decompose any depth-$d\ge2$ multiple polylogarithm/iterated integral into combinations of $Li_4$ terms. Gangl established, via symbol analysis and tensor algebra, that all depth-2 and higher weight-4 iterated integrals admit the following structure (Gangl, 2016):

• Depth 2:

$$I_{3,1}(x,y) + I_{3,1}(y,x) \equiv 0 \text{ mod lower-depth}$$

and

$$I_{3,1}(x,y) \equiv \frac{1}{2}[I_{2,2}(y,x) - I_{2,2}(x,y)] + Li_3(x) Li_1(1/y) + \frac{1}{2} Li_2(x) Li_2(1/y)$$

$I_{2,2}(x, y)$ admits an explicit 9-term $Li_4$ decomposition.

• Depth 3–4:

Any $I_{2,1,1}$ or $I_{1,1,1,1}$ can be written as integer linear combinations of $I_{3,1}$ and $I_{2,2}$, and thus ultimately in $Li_4$.

• Goncharov's Conjecture: For the five-term Bloch group generator $V_0(x,y)$,

$$I_{3,1}(V_0(x,y),z) = \sum_{j=1}^{122} c_j Li_4(f_j(x, y, z)),\;\; c_j \in \{\pm1,\dots,8\}$$

The coefficients and arguments are explicitly determined by linear algebra in tensor spaces and symbol integration. The symbol difference vanishes modulo products, confirming liftability to functions and the completeness of the $Li_4$ functional basis (Gangl, 2016).

3. Shuffle, Stuffle, and General Double-Shuffle Decomposition Formulas

Multiple polylogarithms satisfy deep structural relations arising from concatenation (stuffle) and integral-shuffle algebra:

• General Shuffle Product:

For multi-indices $\mathbf r$, $\mathbf s$ and sequences of variables $\mathbf w$, $\mathbf z$:

$\Li_{\mathbf r}(\mathbf w) \cdot \Li_{\mathbf s}(\mathbf z) = \sum_{(p, \psi) \in J_{k,\ell}} \sum_{\substack{t_1 + \cdots + t_{k+\ell} = |\mathbf r| + |\mathbf s| \ t_i \ge 1}} \left(\prod_{i=1}^{k+\ell} \binom{t_i - 1}{h_{(p, \psi)}(i) - 1}\right) \Li_{\mathbf t}(\mathbf u)$

where $(\mathbf t, \mathbf u)$ run over combinatorial shuffles of orders and argument mergers as determined by $(p, \psi)$, with height functions regulating the indices' placements (Guo et al., 2008).

• Classical Euler Decomposition:

In particular, Euler’s double zeta formula is a specialization:

$$\zeta(r)\zeta(s) = \sum_{k=0}^{r-1} \binom{s+k-1}{k}\zeta(r+k, s-k)+\sum_{k=0}^{s-1} \binom{r+k-1}{k}\zeta(s+k, r-k)$$

and the generalization to arbitrary multiple zeta and multiple polylogarithm products is obtained via the recursive algebra of the shuffle operators.

• Restricted Shuffle Formulas:

Explicit recursive formulas for products of MZVs or $Li$'s with "strings of ones" in the index are fully detailed in (Guo et al., 2014), leading to binomial multi-sum decompositions that generalize Euler’s formula as well as the formula of Eie and Wei. The combinatorics involve multi-indexed summations controlled by composition and partition data.

4. Depth and Variable Reduction Formulas

There exist structural reductions expressing general weight-$n$ multiple polylogarithms in terms of basis elements of lower variable count or lower depth:

• Dan's $n \to n-2$ Reduction: Any weight-$n$ multiple polylogarithm (symbol) with $n$ variables is written as an explicit linear combination of weight-$n$ multiple polylogarithms in at most $n-2$ variables, via affine-invariance and shuffle relations (Dan, 2011). All symbols $[a_0| a_1,\ldots, a_n| a_{n+1}]$ in the Lie coalgebra $H_n(E)$ reduce to combinations built from $R(\alpha,\beta| \gamma,\delta)$ terms, themselves of $n-2$ variables or less. For $n=4$, this gives the “type (3,1)” basis:

$[x,y]^{3,1} = [0\,|\,x, 0, 0, y\,|\,1]$

and all other types (2,2), (1,3) are algebraically reducible to sums of (3,1) and classical 1-variable $Li_4$-type elements.

• Star to Ordinary Polylogarithms: Depth-reduction formulas for star-polylogarithms obtained via Mneimneh-like sum techniques (resolving the Pan–Xu conjecture) yield, for instance,

$\Li_s(a) = \Li_{1, \ldots, 1}^{\star}(1 - p, 1, \ldots, 1 + \frac{ap}{1-p}) - \Li_{1, \ldots, 1}^{\star}(1 - p, 1, \ldots, 1)$

and

$\Li_{1, 1}^\star(x, y) = -\Li_2\Big(1 - \frac{1}{y}\Big)\quad \text{for}\ x = 1-p,\; y = \frac{1}{1-p}$

Such reductions hold at general depths and variable counts, with "Toeplitz limit" methods controlling boundary terms and ensuring vanishing as $n\to\infty$ (Genčev, 2024).

5. Functional Equations and Symmetry-Based Decomposition

Decomposition formulas also arise from involutive symmetries, such as $z \mapsto z/(z-1)$ (“Landen” symmetries), and from $n$-variable affine transformations:

• Landen-Type Decomposition:

For a multi-index $\mathbf k$,

$$Li_{\mathbf{k}}\left(\frac{z}{z-1}\right) + (-1)^{1 + \mathrm{dp}(\mathbf{k})} \sum_{\mathbf{J} \preceq \mathbf{k}} Li_{\mathbf J}(z) = 0$$

where the sum runs over all refinements of $\mathbf{k}$, recovering classical and higher-order Landen-type identities (Shiraishi, 16 Jan 2026).

• Connection-Type and Weighted-Sum Formulas in Two Variables:

Weighted sum formulas and connection formulas for two-variable polylogarithms include identities generalizing the dilogarithm five-term law as well as generating Ohno–Zudilin totals via explicit summations over double indices. These formulas admit specializations to double $T$-values and $L$-values (Kaneko et al., 2024).

6. Coproducts, Symbol Maps, and Hopf Algebra Structure

Multiple polylogarithms are equipped with cocommutative Hopf algebra structures, which provide precise coproduct decompositions:

• Hopf Algebra Coproduct:

For the algebra with generators $[x_1,\ldots,x_d]_{n_1,\ldots,n_d}$, the coproduct $\Delta$ on $Li_{n_1,\ldots,n_d}$ unpacks as a sum over all subwords, reflecting the structure of iterated integrals, e.g.,

$$\Delta Li_{2,1}(x, y) = Li_{2,1}(x, y) \otimes 1 + Li_{1,1}(x, y) \otimes \log x + Li_2(xy) \otimes Li_1(y) + \ldots$$

The maximal deconcatenation map (symbol map) extracts the highest weight part, controlling motivic and functional relations (Greenberg et al., 2022).

• Symbol Map and Variation Matrix:

The symbol map encodes the highest-weight 1-form components, which, together with the variation matrix, represent the mixed Hodge structure variation attached to polylogarithms.

---

This body of decomposition formulas, built from combinatorial shuffle structures, Hopf algebraic principles, and deep symmetry/integration properties, gives a universal framework for the explicit reduction and functional analysis of all iterated polylogarithms, undergirding their application in arithmetic geometry, Feynman diagram evaluation, and motivic cohomology. The explicit expressions at weights up to four are now algorithmic, well-understood, and numerically accessible, with numerous higher-weight, depth, and symmetry-based generalizations found in the current literature (Frellesvig et al., 2016, Gangl, 2016, Guo et al., 2008, Guo et al., 2014, Dan, 2011, Genčev, 2024, Greenberg et al., 2022, Shiraishi, 16 Jan 2026, Kaneko et al., 2024).