Branch Integrals in Mathematical Physics
- Branch integrals are defined by intrinsic branch data that governs the integration family, contour selection, and the geometric nature of resulting periods, as exemplified in dual-conformal integrals in N=4 SYM.
- Fixed-branch representations group propagators to reduce scales and enable efficient high-order ε-expansions in Feynman integrals using dimensional-shift techniques.
- Applications extend to rough-path theory and algebraic geometry, where branch integrals yield explicit formulations for intersection theory, Hodge integrals, and regularized Abelian integrals.
“Branch integrals” is not a single standardized term across mathematics and mathematical physics. In current literature it denotes several distinct but structurally related constructions in which a notion of branch enters the definition of the integral itself: the Coulomb branch of planar SYM, fixed propagator branches in Feynman-integral representations, branch cuts and branch points in analytic continuation, branch maps in Hurwitz and Hodge-theoretic intersection theory, and branched Hopf-algebraic integration in rough-path theory (He et al., 25 Jun 2025, Huang et al., 2024, Afandi, 2019, Bellingeri et al., 2023). The common feature is that the branch datum is not peripheral; it determines either the integral family, the integration contour, the algebraic indexing set, or the geometric meaning of the resulting periods and functions.
1. Coulomb-branch and dual-conformal integrals
In planar SYM, “branch integrals” often refers to the dual-conformal invariant integrals that survive the ten-dimensional lightlike limit of the master correlator and therefore contribute to four-point Coulomb-branch amplitudes. In that setting one writes and , so the condition becomes the on-shell relation . The masses keep the integrals off shell, and the amplitudes are organized into components . The dual-conformal sector contributing to these amplitudes is much smaller than the full conformal-integral sector; the cited analysis lists $1$, $2$, $8$, and 0 relevant DCI integrals at two, three, four, and five loops, respectively. Determinant formulas built from ladder integrals, together with “magic identities,” constrain these Coulomb-branch combinations to all loops (He et al., 13 Feb 2025).
A second development solves infinite families of these four-point Coulomb-branch DCI integrals explicitly. The construction starts from the one-loop box and recursively applies inverse boxing. The resulting functions are parity-odd single-valued harmonic polylogarithms labeled by binary words with no consecutive 1’s; the allowed words are counted by the Fibonacci sequence. Ordinary ladders, generalized ladders compatible with extended Steinmann relations, and the zigzag or anti-prism family all appear as special cases. The same analysis studies their periods, which are generally single-valued multiple zeta values, while the zigzag anti-prism sector gives the classical zigzag periods proportional to 2 (He et al., 25 Jun 2025).
2. Fixed-branch representations and scale reduction
A different usage arises in the theory of fixed-branch integrals. Here propagators are grouped into branches, propagators inside each branch are combined with Feynman parameters, and the resulting objects are fixed-branch integrals 3. Ordinary one-loop Feynman integrals appear as the special case 4. For these objects the dimension-changing transformation gives
5
with the physically useful contour deformation to the negative imaginary 6-axis. This makes high-order 7-expansions efficient because one solves the auxiliary-mass problem in one convenient dimension and reconstructs the target integral by a one-dimensional transform (Huang et al., 2024).
The same branch logic reappears in two other settings. In branch representation for multiloop reduction, propagators with the same quadratic loop-momentum structure are put in one branch, and fixed-branch integrals become the inner basis for intersection-theory reduction. The practical consequence is that 8-loop reduction can be carried out through intersection numbers in at most 9 variables. In multi-propagator angular integrals, iterating the dimensional-shift relation produces branch integrals 0 built from a one-denominator root integral 1; this lowers the number of kinematic scales in the master integrals from 2 to 3. This suggests a common structural role for branch decompositions: they reorganize a multi-scale problem into universal low-scale components indexed by an ordered branching pattern (Huang et al., 6 Apr 2026, Haug et al., 1 Aug 2025).
3. Branch cuts, analytic continuation, and numerical differential equations
In numerical Feynman-integral analysis, branch integrals are integrals whose evaluation is controlled by the branch structure of the differential equation. After reduction to a basis of master integrals, one studies systems such as
4
or more general 5-factorized systems containing non-logarithmic one-forms. The difficulty is that the letters 6 often involve square roots, so the coefficient matrix is multivalued. The operational prescription is to evolve one kinematic variable at a time, factor each active square root as
7
identify the branch points 8, assign branch cuts, and choose a piecewise-linear contour in the complex plane that crosses neither poles nor cuts. Analytic continuation is then encoded geometrically by the contour, not by repeatedly imposing a symbolic 9 rule inside every letter (Rosàs, 5 Mar 2026).
A later implementation pushes this strategy toward Monte Carlo use. Its key simplification is to rewrite a global algebraic root as a product of elementary roots 0, which turns complicated branch-cut geometry into straight rays emanating from explicit branch points. The contour can then be chosen by a practical routing rule, often a horseshoe contour, and branch cuts can be rotated individually when they nearly pinch the available path. In that framework one- and two-loop integral families can be evaluated in double or quadruple precision with runtime scales of milliseconds and hundreds of milliseconds, respectively. A common misconception is that branch handling is secondary to the differential equation itself; these works treat branch selection as part of the numerical definition of the integral (Rosàs et al., 16 Jul 2025).
4. Branch maps, branched coverings, and intersection theory
In algebraic geometry, branch integrals can mean integrals of tautological classes pulled back along a branch map. For the hyperelliptic locus, the branch map
1
and its variant with a conjugate pair lead to the linear hyperelliptic Hodge integrals
2
The paper gives closed formulas for these integrals in terms of elementary symmetric functions of odd or even integers: 3 and
4
These are obtained by Atiyah–Bott localization on stable-map spaces to 5 (Afandi, 2019).
A different branch-point meaning appears in Hurwitz theory. Gaussian integrals over complex matrices with source matrices generate Hurwitz numbers for coverings of arbitrary orientable or non-orientable compact base surfaces, with arbitrary ramification profiles at branch points. In the orientable case the coefficients of trace monomials in source-matrix monodromies are the Hurwitz numbers
6
where the 7 are fixed profiles at “capitals” and the 8 are generated profiles at “watchtowers.” Here “branch” refers literally to branch points of coverings, not to branch cuts of analytic functions (Natanzon et al., 2020).
5. Branched rough paths and non-geometric integration
In rough-path theory, branch integrals are integrals against branched rough paths, meaning characters on the Connes–Kreimer Hopf algebra of rooted forests. The iterated data are indexed by trees and forests rather than by words, so the shuffle identities need not hold and integration by parts may fail. This is the correct algebraic environment for non-geometric rough calculus. On top of this structure one obtains a branched change-of-variable formula: for a solution of a rough differential equation driven by a branched rough path, the increment of a smooth observable is expanded in tree-indexed coefficients via an explicit morphism from the Grossman–Larson Hopf algebra to differential operators. In the semimartingale case this specializes to the classical Itô formula, and the same framework yields a natural Itô–Stratonovich isomorphism by identifying the Connes–Kreimer Hopf algebra with a shuffle algebra over its primitive elements (Bellingeri et al., 2023).
A full integration theorem in this setting constructs the integral of a branched 9-rough path against a one-form of class 0, with 1. The resulting object is again a branched 2-rough path, its 3-variation is quantitatively controlled, and the construction is continuous in rough-path metric. The first level agrees with the first-level integral of the associated 4-rough path. A complementary transfer principle uses Kelly’s bracket extension to define pushforwards of branched rough paths through smooth maps and thereby a coordinate-free integration theory on manifolds once a covariant derivative is fixed (Liu et al., 10 Jan 2026, Ferrucci, 2022).
6. Branch points, principal branches, and regularized Abelian integrals
A further meaning concerns branch-point-dependent special functions. For the sunrise integral and related elliptic Feynman integrals, the branch points of the quartic
5
are the defining data of the elliptic curve. The integration kernels, periods, quasi-periods, and elliptic multiple polylogarithms are all built from these branch points. In this setting the branch points are not an obstruction to be removed; they are the geometric input that determines the correct class of iterated integrals 6 and the kernels 7 (Broedel et al., 2017).
On non-hyperelliptic 8-curves, second kind integrals based at infinity are branch integrals in a literal local sense because infinity is a Weierstrass branch point where all sheets come together. If 9 has a pole at infinity, the paper defines the regularized second kind integral by
0
where 1 is the regularization constant. That constant is a 2-component vector depending on the curve parameters and is uniquely fixed by consistency with the primitive function, sigma-function identities, and Abelian-function relations. Explicit constants are given for the 3, 4, 5, and 6 curves (Bernatska et al., 2017).
At the level of elementary complex analysis, branch-sensitive antiderivatives are organized by explicit principal branches of
7
These branches determine the cut planes and the derivative and antiderivative formulas for the inverse trigonometric and inverse hyperbolic functions. Thus integrals such as
8
become unambiguous only after the branch is fixed. In this literature, “branch integral” refers to an antiderivative whose value, analyticity domain, and continuation across cuts are all determined by a chosen principal branch (Dempsey, 2023).
Across these subjects, the expression is therefore context-dependent rather than uniform. “Branch” may refer to the Coulomb branch, a fixed propagator branch, a branch cut, a branch map, a branch point of a curve or covering, or the branched Hopf-algebraic structure of rough integration. The term’s meaning is controlled entirely by the ambient theory.