Sector Decomposition Method

• Sector Decomposition is an algorithmic method that isolates singularities in multi-dimensional integrals by factorizing UV and IR divergences as explicit poles in ε.
• Its iterative and geometric approaches decompose complex Feynman integrals into regularized sectors, enabling analytic Laurent expansion and high-precision numerical evaluation.
• The method underpins key computational tools like FIESTA and SecDec, playing a pivotal role in advancing high-order perturbative quantum field theory analyses.

Sector decomposition is a constructive algorithmic technique for isolating and extracting the singular structure of multi-dimensional parametric integrals, chiefly those arising in the evaluation of multi-loop Feynman diagrams within dimensional regularization. It provides a systematic method for factorizing ultraviolet (UV) and infrared (IR) divergences as explicit poles in the dimensional regulator ε, thereby reducing the original problem to a finite sum over regularized sector integrals, each manifestly suitable for analytic or numerical computation. The method is fundamental to modern high-order perturbative quantum field theory computations and underpins several general-purpose computational tools.

1. Mathematical Formulation and Divergence Structure

Consider a generic $L$-loop Feynman integral in $D=4-2\epsilon$ dimensions,

$$I(\{\nu_j\}) = \prod_{j=1}^N \int_0^\infty dx_j\, x_j^{\nu_j-1} \, \delta(1-\sum_{i=1}^N x_i) \, U(x)^{a_U(\epsilon)} F(x)^{a_F(\epsilon)}$$

where $U(x)$ and $F(x)$ are Symanzik polynomials encoding graph topology and external kinematics, and $\nu_j$ are the propagator powers. UV and IR divergences correspond to $U(x)\to0$ or $F(x)\to0$ as subsets of Feynman parameters $x_j\to0$, leading to endpoint singularities in parameter space. Dimensional regularization transmutes these divergences into poles in $\epsilon$.

The goal of sector decomposition is to factorize the behavior near all potential singular loci (boundary faces $x_{j_1},\ldots,x_{j_r}\to0$), such that in each sector the integrand takes the form

$$\int_{[0,1]^m} \prod_{j=1}^m dx_j\, x_j^{a_j + b_j\epsilon}\, h(x,\epsilon)$$

where $h(x,\epsilon)$ is smooth and nonvanishing as any $x_j\to0$, exposing all singular behavior in the explicit monomial prefactor. This enables a Laurent expansion in $\epsilon$.

2. Iterated Sector Decomposition Algorithms

The classic iterative approach, pioneered by Binoth–Heinrich and further refined by multiple groups, proceeds in the following structured stages (Tentyukov et al., 2010, Carter et al., 2010, Doncker et al., 2024, Kato, 24 Jan 2026):

1. Primary Sector Decomposition: The simplex constraint $\sum x_j = 1$ is eliminated by partitioning parameter space into $N$ primary sectors, each defined by ordering one variable $x_k$ as the largest. Within sector $k$, set $x_j = x_k t_j$ for $j\neq k$ and $x_k = x_k$, with $0\leq t_j\leq1$.
2. Iterated Subsector Decomposition: Within each sector, examine $U$ and $F$ for remaining zeros at the boundaries. If zeros remain (i.e., overlapping singularities), recursively pick subsets of variables whose simultaneous vanishing causes $U$ or $F$ to vanish, partition the region accordingly, and rescale variables so that each new sector exposes a single variable's leading scaling.
3. Termination: The process is repeated until, in every sector, all $U$ and $F$ are strictly nonzero (finite polynomials) at all boundaries (all $t_j \to 0$), ensuring the singular structure is fully extracted as monomial prefactors.
4. Laurent Expansion and Subtraction: Integrals can then be systematically expanded in $\epsilon$. For any $x_j$ with negative integer power, Taylor expansion and analytic subtraction of endpoint behaviors are performed, yielding explicit residues for $1/\epsilon^k$ poles and finite remainders.

Software packages such as FIESTA (Tentyukov et al., 2010) and SecDec (Carter et al., 2010) automate this algorithm, performing the algebraic decomposition, symbolic subtraction, and subsequent numerical integration using adaptive Monte Carlo methods.

3. Non-Iterative Approaches and Computational Geometry

A significant advance is the geometrical reinterpretation of sector decomposition using Newton polytopes (Kaneko et al., 2010). Given a multivariate polynomial $P(x)$, its singular asymptotics as $x_i\to0$ are controlled by the exponents (multi-indices) of its monomials. Define the Newton polytope $\mathcal{N}(P)$ as the convex hull of these exponent vectors.

• Facets and Sectors: Each facet of $\mathcal{N}(P)$ corresponds to a dominant monomial set controlling asymptotic scaling along certain directions to the origin. The sector structure is thus mapped onto the enumeration of these facets and dual cones in exponent space.
• Algorithmic Steps: The method constructs the convex hull (e.g., via the beneath-beneath algorithm), enumerates all facets and their supporting generators, translates each into a dual cone, and triangulates these into simplicial sectors. Each sector admits a monomial factorization of $P(x)$, so the divergence structure is manifest.
• Empirical Performance: For many diagrams, the non-iterative geometric method produces substantially fewer sectors than iterative decompositions, often by $30$–$75\%$, and avoids pathologies such as infinite decomposition loops (Kaneko et al., 2010).

This geometric approach enhances analytic transparency, aligns sector structure with the combinatorial properties of the integrand, and yields favorable computational scalability for high-loop integrals.

4. Analytic/Numeric Hybrid Strategies and Practical Implementation

In practical applications, sector decomposition is employed for both analytic extraction of divergence structure and high-precision numerical evaluation:

• Pole Extraction: After decomposition, residues of $1/\epsilon^n$ poles are computed as low-dimensional parameter integrals. This can be performed analytically when the relevant functions are sufficiently simple (e.g., for "sunset" and certain three-loop topologies), or semi-analytically by Taylor expansion of the regular factors (Doncker et al., 2024, Kato, 24 Jan 2026).
• Finite Part Evaluation: The finite remainder (the $\epsilon^0$ coefficient) invariably involves smooth integrals over the remaining shape variables. Modern methods employ double-exponential quadrature (Doncker et al., 2024), adaptive Monte Carlo, or deterministic cubature, leveraging parallelism and error estimation from sector-wise computations (Tentyukov et al., 2010, Carter et al., 2010).
• Extrapolation: For extraction of Laurent coefficients, one may numerically evaluate the integrand at several small but finite $\epsilon$ (and, if relevant, $i\rho$ regulators), extrapolating the coefficients via algorithms such as Wynn's $\varepsilon$-algorithm or Richardson extrapolation (Doncker et al., 2024).

A key practical implication is that the number of resulting sectors grows factorially with the number of propagators and loops; for four-loop, 13-propagator cases, the total can reach $\mathcal{O}(10^4)$, and for five loops, up to $\mathcal{O}(10^6)$. This necessitates parallelization (MPI/distributed computation is routine in FIESTA), judicious control over Taylor expansions, and, where possible, analytic simplifications and symmetry reductions (Tentyukov et al., 2010, Carter et al., 2010).

5. Case Studies: Two- and Three-Loop Examples

Recent studies provide detailed worked examples illuminating the concrete realization of sector decomposition:

• Two-loop Sunset Self-energy: The process consists of partitioning into primary sectors, mapping to scale variables ($s, t, u$ or analogs), analytically integrating over the scale variable to expose $1/\epsilon^n$ divergence, and computing the finite part over smooth integrals. For the scalar sunset, up to a $1/\epsilon^2$ UV divergence occurs, with analytic formulae for the pole and finite coefficients in terms of remaining parameter integrals (Kato, 24 Jan 2026).
• Three-loop Two-point Functions: For equal-mass, complete $U$-function graphs, all sectors admit a uniform change of variables yielding a manifest factorization. The Laurent coefficients $C_{-2}$, $C_{-1}$, $C_0$ are computed analytically where possible, finite parts numerically. Energy dependence and physical thresholds (e.g., $s=4m^2, 16m^2$) are transparently analyzed via the sector representation, with threshold singularities managed by double extrapolation in regulator parameters (Doncker et al., 2024).

These studies confirm the robustness of the method for both analytic and pure numerical regimes, with sector numbers and runtime largely determined by the combinatorics of the underlying Feynman graph.

6. Comparative Features and Software Implementations

The two dominant software packages, FIESTA (Tentyukov et al., 2010) and SecDec (Carter et al., 2010), encapsulate the iterative sector decomposition workflow with automated algebraic handling, translation to low-level numeric code (C, Fortran), and parallel adaptive numerical integration (VEGAS/Cuba library). Notable features include:

Package	Language(s)	Decomposition Algorithm	Numeric Backend
FIESTA	Mathematica+C	Iterative+MB options	Cuba (VEGAS/etc)
SecDec	Mathematica+Perl+Fortran	Iterative with heuristics	Cuba/BASES

Both support dimensional regularization, general propagator powers ($a_i+b_i\epsilon$), symbolic subtraction/treatment of endpoint singularities, cluster parallelization, and hybrid analytic–numeric workflows. Sector reduction strategies, parallellization granularity, and convergence accelerators are exposed via user interface or scripting.

A plausible implication is that for integrals at or beyond five-loop complexity, further advances in computational geometry, analytic reduction, and combination with Mellin–Barnes representations will be essential due to the sector proliferation and numerical cost.

7. Outlook and Extensions

Sector decomposition has become an essential pillar in the practical calculation of high-order quantum corrections where analytic results are intractable or unavailable. Its centrality in modern applications such as precision collider physics, effective field theory, and beyond-standard-model computations is established. Emerging non-iterative algorithms based on convex geometry are poised to further reduce computational complexity and minimize the number of sector integrals, suggesting a research focus on hybrid, geometry-guided sector reduction, symbolic–numeric integration, and further parallel acceleration (Kaneko et al., 2010).

The method also finds application beyond pure loop integrals: in phase-space integrations, phase transitions for diagrammatic Monte Carlo, and certain classes of algebraic singularities. As computational resources and algorithmic innovations advance, sector decomposition’s systematic, automatable treatment of divergences remains indispensable for precision multi-loop quantum field theory.