KLT Double Copy in Gauge and Gravity Theories

• KLT double copy is a framework that expresses gravitational amplitudes as quadratic combinations of gauge-theory amplitudes via an explicit KLT kernel.
• Its algebraic structure leverages color/kinematics duality and monodromy relations, generalizing to higher-derivative and effective field theory setups with minimal-rank constraints.
• Extensions of the framework include massive states, loop-level integrands, and a geometric interpretation via twisted (co)homology, unifying diverse amplitude constructions in string and field theories.

The KLT double copy is a set of algebraic relations between scattering amplitudes in gauge, gravity, and string theory, organizing the construction of gravitational (or closed-string) amplitudes as quadratic combinations of gauge-theory (or open-string) amplitudes via a distinguished kernel. It originates from the work of Kawai, Lewellen, and Tye, who derived explicit integral-level correspondences between open- and closed-string amplitudes; over the past decades, this structure has been bootstrapped, generalized, and abstracted to a universal organizing principle underlying a broad class of tree-level and loop-level (super)gravity, string, and effective field theories. The central object is the KLT kernel, which encodes both the color/kinematics duality of the gauge side and the single-valued transcendental structure of the gravitational/closed side.

1. Algebraic Structure and KLT Kernel

The foundational operation of the KLT double copy is the mapping

$$\mathcal{M}_n = \sum_{\alpha,\beta} A_n^L[\alpha] \, S_n[\alpha|\beta] \, A_n^R[\beta]$$

where $A_n^{L,R}$ are color-ordered tree-level amplitudes of two factor theories (such as Yang–Mills), and $S_n[\alpha|\beta]$ is the KLT (momentum) kernel—an explicitly computable function of kinematic invariants. For pure gauge theory (biadjoint scalar) at tree level, $S_n$ is the inverse of the matrix of doubly color-ordered scalar amplitudes $m_n$ evaluated in a minimal $\operatorname{BCJ}$ (Kleiss–Kuijf/BCJ) basis of size $(n-3)!$ (Chi et al., 2021). The KLT multiplication $\otimes$ is associative, has the identity “zeroth copy” (biadjoint scalar), and obeys

$L \otimes 1 = L \,,\quad 1 \otimes R = R$

for any color-ordered theory.

In string theory, the KLT kernel is deformed to include $\alpha'$ dependence,

$A_n^{L,R}$0

where every propagator $A_n^{L,R}$1 is replaced by $A_n^{L,R}$2 or $A_n^{L,R}$3, generating an infinite tower of higher-derivative corrections (contact/UV-completion effects).

Higher-Derivative and EFT Generalizations: In effective field theories (EFTs), the double copy persists if one admits a family of generalized KLT kernels consistent with minimal rank, locality, and factorization constraints. At four points, the most general double copy kernel compatible with these properties is constructed in terms of two parameters (interpreted as effective $A_n^{L,R}$4s), and a symmetric function $A_n^{L,R}$5 encoding higher-derivative freedom (Chen et al., 2023, Chi et al., 2021, Bonnefoy et al., 2021). Explicitly,

$A_n^{L,R}$6

with $A_n^{L,R}$7 a “seed” function and $A_n^{L,R}$8 a sum over higher-dimension operators, each coefficient fixed by the minimal-rank and crossing constraints.

Table: Summary of KLT Kernel Forms

Theory	KLT Kernel Structure	Notes
Field theory	Rational in $A_n^{L,R}$9	Inverse of $S_n[\alpha|\beta]$0
String theory	Trigonometric ($S_n[\alpha|\beta]$1, $S_n[\alpha|\beta]$2)	$S_n[\alpha|\beta]$3-deformed
EFT General	2 free slopes + $S_n[\alpha|\beta]$4	Higher-derivative params

2. Generalizations: Higher Derivative, Heterotic, and Form Factors

Allowing distinct left/right single-copy amplitudes $S_n[\alpha|\beta]$5 and $S_n[\alpha|\beta]$6, subject to generalized monodromy or BCJ/Kleiss–Kuijf relations, admits a wide array of “heterotic-type” double copies, with the possibility of distinct color structures (e.g., inclusion of $S_n[\alpha|\beta]$7 symmetric tensors in the color algebra, unlike pure Yang–Mills, which involves only antisymmetric $S_n[\alpha|\beta]$8) (Chi et al., 2021).

• EFT Bootstrap: The solution space of generalized KLT double copies is fixed order by order via minimal rank and locality at four, five, and higher points. For four points, essentially only $S_n[\alpha|\beta]$9 and “even-derivative” coefficients $S_n$0 remain free after imposing six-point constraints (Chen et al., 2023). Thus, two effective slopes (interpretable as “left” and “right” $S_n$1s) parametrize all consistent deformations.
• Form Factors: The KLT double copy has been extended to gauge-invariant operator insertions (form factors of $S_n$2 or $S_n$3), introducing new operator-induced color identities. The associated KLT kernels admit a factorization structure in terms of universal $S_n$4 vectors, and the explicit cubic-graph basis (CK-dual) numerators are constructed to match all generalized Jacobi and operator identities (Lin et al., 2023).

3. Twisted (Co)homology and Intersection Interpretation

The KLT kernel, both in string and field theory, admits a deep geometric interpretation in terms of twisted (co)homology and intersection theory on moduli spaces of punctured spheres (genus $S_n$5) or tori (genus $S_n$6).

• Sphere/Genus $S_n$7: At genus zero, the intersection numbers of twisted cycles (integration contours weighted by the Koba–Nielsen factor) provide the canonical basis in which the KLT double copy is manifest (Britto et al., 2021). The double copy formula becomes

$S_n$8

where the KLT kernel is the inverse intersection matrix of appropriately chosen cycles, and the “single-valued map” links open- to closed-string periods.

• Torus/Genus $S_n$9 and AdS: The genus-one situation generalizes: twisted homologies on the once-punctured torus, along with intersection and index matrices, define a genus-one KLT-like kernel. The closed-string integral is obtained by sewing together two open-string periods using the kernel constructed from intersection indices (Bhardwaj et al., 2023). In AdS, noncommutative twisted de Rham theory encapsulates the building-block double copy structure, with the kernel given by the inverse of a (noncommutative) intersection number, enabling explicit inversion at four points (Kakkad et al., 29 Dec 2025).

4. Massive Double Copy, Kaluza-Klein, and Constraints

For massive external states, the KLT double copy is modified by a mass-sensitive kernel, which is the inverse of the matrix of massive biadjoint scalar amplitudes. The physicality of the double copy requires not only color-kinematics duality but also additional spectral constraints—generically, the rank of the biadjoint matrix must remain minimal for spurious singularities to be avoided (Johnson et al., 2020, Li et al., 2022, Li et al., 2021).

• Kaluza-Klein (KK) States: Upon toroidal compactification, the mass spectrum of KK states satisfies a group-theory constraint (KK-number conservation), ensuring the existence of a physically sensible massive KLT double copy (Li et al., 2022). The general $m_n$0-point amplitude for massive KK gravitons is assembled as a sum over products of orderings of massive gauge amplitudes, with the kernel sensitive to the KK charges and momenta.
• Universality and Spectral Equations: Only specific mass assignments (as in compactified string/field theory) permit a spurious-pole-free KLT/double copy structure. In generic massive Yang–Mills or arbitrary mass spectra, the five- and higher-point KLT kernel can develop spurious singularities, and the construction can fail to yield physical amplitudes unless rank conditions are met (Johnson et al., 2020).

5. Loop-Level Extensions and Field-Theoretic Realizations

At one loop and beyond, KLT-type double-copy relations persist at the level of integrands. The kernel at one loop is the inverse of the biadjoint scalar one-loop partial amplitude matrix, up to the appropriate basis and combinatorics (Cao et al., 29 Sep 2025). In string theory, closed-string one-loop amplitudes are organized as loop-momentum integrals of products of open-string amplitudes, with an elliptic splitting/monodromy kernel (Stieberger, 2023).

• Forward-Limit and Recursion: The field-theoretic construction uses forward limits and “single cut” methods to reconstruct loop integrands from tree-level blocks, applying KLT at the level of unitarity cuts and respecting locality and gauge invariance. For each $m_n$1-point one-loop gravity integrand,

$m_n$2

where $m_n$3 is the loop-level KLT kernel (Cao et al., 29 Sep 2025).

• Classical/Off-Shell Double Copy: The double copy extends to off-shell Berends–Giele currents for gravity (not just amplitudes), including all NS-NS (and by extension heterotic and RR) sectors, with the KLT kernel continuing to glue left and right sectors up to gauge/polynomial shifts (Cho et al., 2021).
• Twistor Space Realizations: The KLT kernel also arises naturally in twistor space formulations, particularly in the helicity-graded S-matrices of $m_n$4 supergravity, where it glues together maximally helicity-violating (MHV) blocks via a graph-theoretic interpretation and admits closed expressions in twistor moduli (Adamo et al., 2024).

6. Physical Implications and Single-Valued Projections

A central implication of the KLT double copy is the emergence of the “single-valued projection” at the level of transcendental constants: in the low-energy expansion, the double copy removes all even Riemann $m_n$5-values (e.g., $m_n$6) and doubles all odd ones ($m_n$7) (Chen et al., 2023). Closed-string amplitudes thus inherit this property by construction, matching the single-valued map in motivic cohomology and polylogarithmic expansions.

In summary, the KLT double copy forms the backbone of a universal web of relations between gauge-theory, gravity, and string amplitudes, controlling both their algebraic structure and transcendental content. Its kernel is rigidly constrained by minimal rank, locality, and global monodromy/BCJ relations, yet admits a rich deformation space encoding all consistent higher-derivative, massive, and curved-space generalizations currently known (Chen et al., 2023, Chi et al., 2021, Britto et al., 2021, Bhardwaj et al., 2023, Kakkad et al., 29 Dec 2025).